BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:TIFR Cafeteria and Related Queues - A Game of Arrivals
DTSTART;VALUE=DATE-TIME:20091111T113000Z
DTEND;VALUE=DATE-TIME:20091111T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-230@cern.ch
DESCRIPTION:In this talk\, we introduce the `cafeteria or the banquet hall
queuing problem' strongly motivated by speaker's experience of waiting in
the TIFR cafeteria queue: Fixed but a large number of users arrive into a
queue which provides service starting at a fixed time\, say\, 12:15 pm.
Users may (and some do) arrive before this time and queue up. Their cost
is a function of their waiting time in the queue and time at which service
is received. We analyze this system in an asymptotic regime and develop f
luid limit for the resultant queuing system. The limiting system may be
modeled as a non-atomic game for which we determine the Nash-Wardrop equil
ibrium arrival strategy under a variety of assumptions on the cost structu
re. Furthermore\, we note that the `price of anarchy' of this system equal
s 2 under linearity and homogeneity assumptions. We discuss some potential
ways to limit this\nanarchy.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=230
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=230
END:VEVENT
BEGIN:VEVENT
SUMMARY:Counting the Number of Spanning Trees in a Graph
DTSTART;VALUE=DATE-TIME:20091113T103000Z
DTEND;VALUE=DATE-TIME:20091113T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-237@cern.ch
DESCRIPTION:A classic counting algorithm\, Kirchhoff's theorem gives a for
mula for finding the number of spanning trees in a simple\, connected\, un
directed graph. I will discuss a simple proof of this theorem\, based on e
lementary linear algebra.\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=237
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=237
END:VEVENT
BEGIN:VEVENT
SUMMARY:#Planar-PM in Poly-time
DTSTART;VALUE=DATE-TIME:20091120T103000Z
DTEND;VALUE=DATE-TIME:20091120T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-249@cern.ch
DESCRIPTION:The number of perfect matchings in planar graphs can be comput
ed in polynomial (in the number of vertices) time.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=249
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=249
END:VEVENT
BEGIN:VEVENT
SUMMARY:Turan's Graph Theorem
DTSTART;VALUE=DATE-TIME:20091127T103000Z
DTEND;VALUE=DATE-TIME:20091127T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-263@cern.ch
DESCRIPTION:Consider the set of all n-vertex graphs that does not contain
a k-clique. What is the maximum number of edges that any graph in this set
have?\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=26
3
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=263
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Quantum f-relative Entropy and Generalized Data Processing
Inequalities
DTSTART;VALUE=DATE-TIME:20091208T103000Z
DTEND;VALUE=DATE-TIME:20091208T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-266@cern.ch
DESCRIPTION:We study the fundamental properties of the quantum f-relative
entropy\, where f(.) is an operator convex function. We give the equality
conditions under various properties including monotonicity and joint conve
xity\, and these conditions are more general than\, since they hold for a
class of operator convex functions\, and different for f(t) = -log(t) from
\, the previously known conditions. The quantum f-entropy is defined in te
rms of the quantum f-relative entropy and we study its properties giving t
he equality conditions in some cases. We then show that the f-generalizati
ons of the Holevo information\, the entanglement-assisted capacity\, and t
he coherent information also satisfy the data processing inequality\, and
give the equality conditions for the f-coherent information.\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=266
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=266
END:VEVENT
BEGIN:VEVENT
SUMMARY:Scheduling Multi-Antenna Broadcast Systems with Heterogeneous User
s
DTSTART;VALUE=DATE-TIME:20091210T060000Z
DTEND;VALUE=DATE-TIME:20091210T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-270@cern.ch
DESCRIPTION:Information-theoretic results indicate that multi-user multi-a
ntenna systems provide potentially huge capacity gains. However\, these r
esults assume sophisticated coding techniques and perfect channel state in
formation which may require a prohibitive amount of feedback. We show that
transmitting to just a few suitably selected users is asymptotically opti
mal as the user population grows large\, while dramatically reducing the f
eedback overhead and operational complexity. Specifically\, we consider a
multi-antenna broadcast system with M transmit antennas\, and a heterogene
ous user population. we tackle the problem of maximizing a weighted sum ra
te quantity. We establish a novel upper bound for the weighted sum capacit
y\, which we then use to show that the maximum expected weighted sum rate
can be asymptotically achieved by transmitting to a suitably selected subs
et of at most MC users\, where C denotes the number of distinct user class
es. Numerical experiments indicate that the asymptotic results are remarka
bly accurate and that the proposed schemes operate close to absolute perfo
rmance bounds\, even for a moderate number of users.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=270
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=270
END:VEVENT
BEGIN:VEVENT
SUMMARY:SHRUTI: An Industrial Strength Formally Certified SAT Solver
DTSTART;VALUE=DATE-TIME:20091210T090000Z
DTEND;VALUE=DATE-TIME:20091210T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-286@cern.ch
DESCRIPTION:Boolean satisfiability (SAT) solvers are not only routinely us
ed in the formal verification of large industrial problems\, but also appl
ied in safety-critical domains such as the railways\, avionics\, and autom
otive industries. However\, the use of SAT solvers in such domains require
s some form of assurance for the results\, as the solvers can have bugs. T
he complexity of modern\, highly optimized SAT solvers makes it very diffi
cult to provide a formal proof of correctness.\n\nIn this talk I will pres
ent a new approach for certifying SAT solvers where an un-trusted\, indust
rial-strength\, SAT solver is plugged into a trusted\, formally certified\
, SAT proof checker to provide an industrial-strength formally certified S
AT solver --- SHRUTI. The three key novelties of our approach are:\n\n1. O
ur checker is formally designed and proven correct in an LCF style proof a
ssistant and is automatically extracted from the proof assistant.\n\n2. It
is used as a standalone executable program independent of any supporting
theorem prover\, and\n\n3. It certifies any SAT solver respecting the agre
ed format for satisfiable and unsatisfiable claims.\n\nThe main part of ou
r work involves implementing a certified proof checker for unsatisfiable c
laims that is formally designed\, verified\, and implemented inside Coq. I
will present its design and outline the correctness aspects in the talk\,
and show evaluation results of SHRUTI on a representative set of industri
al benchmarks from the SAT Race Competition. The performance results demon
strate that whilst SHRUTI is slower (~2.5 times) than the uncertified solv
ers PicoSAT/Tracecheck and zChaff/ZVerify\; it is faster (upto 32 times) t
han the previously developed certified checker implemented on top of the H
OL 4 theorem prover.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=286
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=286
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Designs (or Packings)
DTSTART;VALUE=DATE-TIME:20091211T093000Z
DTEND;VALUE=DATE-TIME:20091211T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-298@cern.ch
DESCRIPTION:I will give 3 proofs for showing\,\n\nThere exists a family of
subsets \\mathfrak{F}\, of \\{ \\cdots n\\}\,\n\n * such that each ele
ment A \\in \\mathfrak{F} is of size \\frac{n}{4}\n * for any pair A\,B
\\in \\mathfrak{F}\,\\ |A \\cap B| \\leq \\frac{n}{8}\,\n * and |\\mat
hfrak{F}| = 2^{\\Omega(n)}\n\nhttps://indico.tifr.res.in/indico/conference
Display.py?confId=298
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=298
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction to MetaMathematics
DTSTART;VALUE=DATE-TIME:20091218T093000Z
DTEND;VALUE=DATE-TIME:20091218T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-341@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
341
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=341
END:VEVENT
BEGIN:VEVENT
SUMMARY:Large Deviations Of Max-Weight Scheduling Policies
DTSTART;VALUE=DATE-TIME:20091221T053000Z
DTEND;VALUE=DATE-TIME:20091221T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-267@cern.ch
DESCRIPTION:We consider a single-server discrete-time system with $K$ user
s where the server picks operating points from a compact\, convex and coor
dinate convex set in $\\Re_+^K$. For this system we analyse the performanc
e of a stablising policy that at any given time picks operating points fro
m the allowed rate region that maximise a weighted sum of rate\, where the
weights depend upon the workloads of the users. In particular\, we are in
terested in a Large Deviations based analysis of this policy\, and under b
oth the ``large-buffer" and ``many-sources" regimes. The unifying theme of
this work is to prove a Large Deviations Principle (LDP) for the queueing
process using an appropriate generalization of the contraction principle\
, namely\, Puhalskii's extended contraction principle and Garcia's extende
d contraction principle.\n\nhttps://indico.tifr.res.in/indico/conferenceDi
splay.py?confId=267
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=267
END:VEVENT
BEGIN:VEVENT
SUMMARY:Protecting Sensitive Information from Untrusted Code
DTSTART;VALUE=DATE-TIME:20091221T090000Z
DTEND;VALUE=DATE-TIME:20091221T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-285@cern.ch
DESCRIPTION:As computer systems support more aspects of modern life\, from
finance to health care\, security becomes increasingly important. Compute
r systems have to safeguard data from unauthorized users\, malicious code\
, code with programmer bugs and other threats.\n\nRecently\, decentralized
information flow control (DIFC) has emerged as a promising model to write
programs with powerful\, end-to-end security guarantees. The DIFC model p
rovides security by allowing users to associate secrecy and integrity labe
ls with data and restricting the flow of information according to these la
bels.\n\nCurrent DIFC systems that run on commodity hardware can be broadl
y categorized into two types: language-level and operating system-level DI
FC. Language level solutions provide no guarantees against security violat
ions on system resources\, like files and sockets. Operating system soluti
ons can mediate accesses to system resources\, but are inefficient at moni
toring the flow of information through fine-grained program data structure
s. \n\nIn the talk I will primarily describe "Laminar"\, the first system
to implement decentralized information flow control using a single set of
abstractions for OS resources and heap-allocated objects. Programmers expr
ess security policies by labeling data with secrecy and integrity labels\,
and then access the labeled data in lexically scoped security regions. La
minar is implemented using a modified Java virtual machine and a new Linux
security module.\n\nI will also give a brief overview of another system t
hat I am developing\, called "Airavat". Airavat is a MapReduce-based syste
m that provides strong security and privacy guarantees for distributed com
putations on sensitive data.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=285
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=285
END:VEVENT
BEGIN:VEVENT
SUMMARY:Wireless Scheduling with Limited Information
DTSTART;VALUE=DATE-TIME:20091223T103000Z
DTEND;VALUE=DATE-TIME:20091223T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-373@cern.ch
DESCRIPTION:We consider the problem of scheduling with incomplete and/or l
ocal information in wireless environments. Wireless systems typically suff
er from channel fading and interference\, and good scheduling algorithms a
re necessary to adapt to and mitigate these effects. Classical studies on
wireless scheduling investigate in much detail settings where the full sta
te of the system is available to the scheduler while making scheduling dec
isions. In contrast\, we focus on cases where valuable network state infor
mation is lacking at the scheduler\, and study its resulting effect on sys
tem performance. Insights gained from the analysis are used to develop eff
ective wireless scheduling algorithms that can operate in a limited inform
ation context\, at the same time exploiting the full capabilities of the s
ystem to yield good performance - captured by metrics like throughput and
delay. We present results on the throughput region and throughput-optimal
algorithms for two representative wireless downlink scheduling problems wh
ich model restricted channel state information and limited coordination ca
pabilities (joint work with Constantine Caramanis and Sanjay Shakkottai).\
n\nBio: Aditya Gopalan is a fourth year PhD student at the Department of E
lectrical and Computer Engineering\, The University of Texas at Austin. He
received his Bachelors and Masters degrees in Electrical Engineering from
the Indian Institute of Technology Madras in 2006. His research interests
include wireless networks\, scheduling and queueing theory.\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=373
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=373
END:VEVENT
BEGIN:VEVENT
SUMMARY:Universal Coding Theory in Quantum Systems
DTSTART;VALUE=DATE-TIME:20091224T103000Z
DTEND;VALUE=DATE-TIME:20091224T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-379@cern.ch
DESCRIPTION:We construct a universal code for stationary and memoryless cl
assical-quantum channel as a quantum version of the universal coding by Cs
isz\\'{a}r and K\\"{o}rner. Our code is constructed by the combination of
irreducible representation\, the decoder introduced through quantum inform
ation spectrum\, and the packing lemma. Also\, we have proven that there e
xists a quantum state approximating any multi-copy state universally when
we measure the error by means of the normalized relative entropy. While th
e qubit case was proven by Krattenthaler and Slater\, the general case has
been open for more than ten years. For a deeper analysis\, we have solved
the mini-max problem concerning `approximation error' up to the second or
der. Furthermore\, we have applied this result to quantum lossless data co
mpression\, and have constructed a universal quantum lossless data compres
sion.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=379
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=379
END:VEVENT
BEGIN:VEVENT
SUMMARY:Capacity Planning in Service Systems with Arrival Rate Uncertainty
: Safety Staffing Principles Revisited
DTSTART;VALUE=DATE-TIME:20100105T060000Z
DTEND;VALUE=DATE-TIME:20100105T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-380@cern.ch
DESCRIPTION:We study a capacity sizing problem in service systems with unc
ertain arrival rates\; telephone call centers are canonical examples of su
ch systems. The objective is to choose a staffing level that minimizes the
sum of personnel costs and abandonment/waiting time costs. We formulate
a simple fluid analogue\, which is in essence a newsvendor problem\, and d
emonstrate that the solution it prescribes performs remarkably well. In pa
rticular\, the gap between the performance of the optimal staffing level a
nd that of our proposed prescription is independent of the ``size" of the
system\, i.e.\, it remains bounded as the system size (demand volume) incr
eases. This stands in contrast to the more conventional theory that applie
s when arrival rates are known\, and commonly used rules-of-thumb predicat
ed on it. Specifically\, in that setting the difference between the optima
l performance and that of the fluid solution diverges at a rate proportion
al to square-root of the size of the system. One manifestation of this is
the celebrated square root safety staffing principle that dates back to wo
rk of Erlang\, which augments solutions of the deterministic analysis with
additional servers of order square root the volume of demand. In our work
\, we establish that this type of prescription is needed only when arrival
rates are suitably ``predictable."\n\nhttps://indico.tifr.res.in/indico/c
onferenceDisplay.py?confId=380
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=380
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimal Policies for Large Queueing Networks Through Diffusion Lim
its
DTSTART;VALUE=DATE-TIME:20100108T060000Z
DTEND;VALUE=DATE-TIME:20100108T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-335@cern.ch
DESCRIPTION:Queueing networks are ubiquitous\, serving as models of comput
er networks\, service stations\, dams\, etc. Their performance depends on
their structure\, parameters and on the service rules. In general\, findin
g good service rules is a difficult problem. We give some examples to moti
vate the study of queueing networks\, and in particular large networks. We
show how to approximate large networks by simpler models of diffusions. T
wo examples will then illustrate how the asymptotic analysis can lead to p
ractical conclusions about service rules\, when the model is not known com
pletely---in this case\, service rates of the various servers are not know
n. The first example shows that\, when attempting to minimize the time in
the system\, it suffices to take a surprisingly small sample of service ti
mes of a subset of servers. The second example deals with the issue of fai
rness towards servers\, and yields a rule which does not depend on the val
ues of service rates\, and is easy to implement as well as robust with res
pect to some model assumptions (joint work with Rami Atar and Yair Shaki).
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=335
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=335
END:VEVENT
BEGIN:VEVENT
SUMMARY:What's Decidable for Asynchronous Programs?
DTSTART;VALUE=DATE-TIME:20100114T053000Z
DTEND;VALUE=DATE-TIME:20100114T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-407@cern.ch
DESCRIPTION:An asynchronous or ``event-driven" program is one that contain
s procedure calls which are not directly executed from the callsite\, but
stored and executed later by an external scheduler. By providing a low-ove
rhead way to manage concurrent interactions\, asynchronous programs form t
he core of many server programs\, embedded systems\, and popular programmi
ng styles for the web (Ajax). Unfortunately\, such programs can be hard to
write and maintain as sequential control flow needs to be split into seve
ral disjoint handlers. They are a challenge for static analysis tools as t
hey are infinite state: both the program stack and the number of outstandi
ng asynchronous requests may be unbounded. We show that safety and livenes
s properties can be checked effectively for the class of Boolean asynchron
ous programs\, thus enabling automatic static techniques to check for corr
ectness or for errors.\n\nBio: Rupak Majumdar is an Associate Professor in
the Department of Computer Science at the University of California\, Los
Angeles and Scientific Director at the Max Planck Institute for Software S
ystems. His research interests are in the design and verification of react
ive hardware and software systems. He received the President's Gold Medal
from IIT\, Kanpur\, the Leon O. Chua award from UC Berkeley\, and an NSF C
AREER award.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=407
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=407
END:VEVENT
BEGIN:VEVENT
SUMMARY:Refinement of the Generalized Nash Equilibrium
DTSTART;VALUE=DATE-TIME:20100120T103000Z
DTEND;VALUE=DATE-TIME:20100120T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-383@cern.ch
DESCRIPTION:We are concerned with generalized Nash games in which the play
ers' strategy sets are coupled by a shared constraint. A widely employed s
olution concept for these games is the generalized Nash equilibrium (GNE)\
, which is the natural generalization of the Nash equilibrium. The variati
onal equilibrium (VE) is a specific kind of GNE given by a solution of the
variational inequality formed from the common constraint and the mapping
of the gradients of player objectives. Our contribution is in providing co
nditions under which the existence of a GNE is necessary and sufficient fo
r the existence of a VE\; in such an instance\, the VE is said to be a ``r
efinement" of the GNE. Establishment of this result is seen to be of relev
ance to pure\, applied and computational game theorists. We present a theo
ry that gives sufficient conditions for the VE to be a refinement of the G
NE. For certain games these conditions are shown to be necessary. This the
ory rests on a result showing that the GNE and the VE are equivalent upto
the Brouwer degree of certains functions whose zeros are the GNE and VE re
spectively. In the primal space we show equality between the Brouwer degre
es of the natural maps of the quasi-variational inequality and the variati
onal inequality\, whose solutions are the GNE and VE respectively. Using a
novel equation reformulation of the VE\, this result is extended to the p
rimal-dual space. These degree theoretic relationships pave the way for th
e aforesaid sufficient conditions. Our results unify some known results ab
out shared constraint games and provide mathematical justification for ide
as that were known to be appealing to economic intuition.\n\nBio: Ankur is
a doctoral student at the University of Illinois at Urbana-Champaign (UIU
C)\, USA. He received his B.Tech. in Aerospace Engineering from Indian Ins
titute of Technology\, Bombay in 2006 and M.S. in General Engineering from
UIUC in 2008. He was a visiting scholar at the Tata Institute of Fundamen
tal Research\, Mumbai in 2008. His interests are in the theory of games\,
economics\, mathematical programming and applied probability. His work has
been based on providing topological insights into equilibria of generaliz
ed Nash games and multi-leader-multi-follower games.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=383
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=383
END:VEVENT
BEGIN:VEVENT
SUMMARY:Option Pricing and Hedging in the Incomplete Market
DTSTART;VALUE=DATE-TIME:20100121T113000Z
DTEND;VALUE=DATE-TIME:20100121T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-340@cern.ch
DESCRIPTION:An important aspect of the stock price process\, which has oft
en been ignored in the financial literature\, is that prices on organized
exchanges are restricted to lie on a grid. We consider continuous-time mo
dels for the stock price process with random waiting times of jumps and di
screte jump size. We consider a class of pure jump processes that are ``cl
ose'' to the Black-Scholes model in the sense that as the jump size goes t
o zero\, the jump model converges to geometric Brownian motion. We study
the changes in pricing caused by discretization. Upper and lower bounds o
n option prices are developed. We show that it is possible to hedge option
s if one restricts to jumps of size one\, that is\, if one models the stoc
k price process as a birth and death process. One needs the stock and anot
her market traded derivative to hedge an option in this setting. We obtain
parameter estimates using Generalized Method of Moments. We use filtering
equations for inference in the stochastic intensity setting. We present r
eal data applications to study the performance of our modeling and estima
tion techniques.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=340
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=340
END:VEVENT
BEGIN:VEVENT
SUMMARY:Comparison Theorems and the Validity of Heavy Traffic Limit Distri
butions for Stochastic Networks
DTSTART;VALUE=DATE-TIME:20100125T103000Z
DTEND;VALUE=DATE-TIME:20100125T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-394@cern.ch
DESCRIPTION:The issue of stability of stochastic networks continues to be
one of the central issues in applications along with the computation of th
e stationary distribution. Under heavy traffic conditions\, in recent work
s by Gamarnik and Zeevi (AAP'06) and Budhiraja and Lee (MOR'09) they showe
d that the steady state limits of networks with i.i.d inputs that satisfy
a FCLT converge to the corresponding stationary distribution of the limit
SRBM (Semi-martingale Reflecting Brownian Motion) or in other words the la
rge time limit and the scaling limit can be interchanged. Their arguments
were based on Lyapunov techniques.\n\nIn this talk we consider networks wi
th Levy inputs and show that the corresponding interchange result holds vi
a direct arguments without the need for Lyapunov techniques. Indeed the re
sults follow in a direct manner from some well known results on stochastic
networks with Levy inputs due to Kella and Whitt.\n\nIn the second part o
f the talk I will show some new comparison theorems for stochastic network
s that yield useful results for showing conditions when monotonicity will
hold. We conclude with a generalization of these results to general reflec
ted diffusions with jumps even allowing for state-dependent reflections wh
en the continuity of the Skorohod map cannot be shown (joint work with Dr.
Francisco Piera (University of Chile) and Jean-Paul Haddad (Waterloo)).\n
\nBio: The speaker was educated at the Indian Institute of Technology\, Bo
mbay (B.Tech\, 1977)\, Imperial College\, London (MSc\, DIC\, 1978) and UC
LA (PhD\, 1983). He is currently a University Research Chair Professor in
the Dept. of ECE at the University of Waterloo\, Ont.\, Canada where he ha
s been since September 2004. Prior to this he was Professor of ECE at Purd
ue University\, West Lafayette\, USA where he continues to be an Adjunct P
rofessor. He is a Fellow of the IEEE and the Royal Statistical Society. He
is a recipient of INFOCOM 2006 Best Paper Award and was runner-up for the
Best Paper Award at INFOCOM 1998. He has served as an editor of the IEEE/
ACM Trans on Networking (2004-09) and as guest editor for a number of spec
ial issues of networking and applied probability related journals. He is a
lso the author of a forthcoming monograph on performance modeling statisti
cal multiplexing in networks to be published by Morgan and Claypool in ear
ly 2010. His research interests are in modeling\, control\, and performanc
e analysis of both wireline and wireless networks\, and in applied probabi
lity and stochastic analysis with applications to queueing\, filtering\, a
nd control.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=394
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=394
END:VEVENT
BEGIN:VEVENT
SUMMARY:Kilim: An Actor Framework for Servers
DTSTART;VALUE=DATE-TIME:20100125T090000Z
DTEND;VALUE=DATE-TIME:20100125T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-398@cern.ch
DESCRIPTION:Servers are always part of a distributed ecosystem. Less focus
is given to the fact that servers are also deployed on internally distrib
uted hardware\, namely multicore/ multiprocessor machines. Shared-memory c
oncurrency -- the current approach -- is extremely error-prone and unmaint
ainable. One alternative is to program with communicating actors\, which a
re active objects with a private heap and thread of their own\; they do no
t share memory and communicate and synchronize only by passing messages.\n
\nMy research has focussed on making actor-oriented programming viable and
performance-competitive in extreme server environments such as Amazon.com
or financial systems. The solution is part of the Kilim framework (http:/
/kilim.malhar.net). We find it to be much faster and scalable than existin
g best-of-industry approaches\; we will look at the Jetty web server and t
he Berkeley DB database\, and compare them to Kilim variants.\n\nThe talk
will provide an overview of two aspects of my research: an annotation-base
d type system to control pointer aliasing (to obtain isolation and for zer
o-copy messaging)\, and a continuation-passing style transformation of Jav
a bytecode to obtain ultra-lightweight threads (can have hundreds of thous
ands of threads).\n\nBio: Sriram Srinivasan has been designing and deliver
ing systems for the past 22 years. He has worked on high-performance trans
actional systems\, CORBA and EJB frameworks\, flight planning systems and
communication protocols\, to name a few. He was one of the primary designe
rs and engineers of the Weblogic Application server. He recently took a de
tour via the University of Cambridge for a PhD in computer science.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=398
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=398
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dryad and DryadLINQ
DTSTART;VALUE=DATE-TIME:20100127T090000Z
DTEND;VALUE=DATE-TIME:20100127T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-435@cern.ch
DESCRIPTION:This talk describes a set of distributed services developed at
Microsoft Research Silicon Valley to enable efficient parallel programmin
g on very large datasets. Parallel programmes arise naturally within scien
tific\, data mining\, and business applications. Central to our philosophy
is the notion parallel programmes do not have to be difficult to write an
d that the same programme must seamlessly run on a laptop\, desktop\, a sm
all cluster\, or on a large data center without the author having to worry
about the details of parallelization\, synchronization\, or fault-toleran
ce. Dryad and DryadLINQ are two services that embody this belief. The comb
ination is extensively used within Microsoft\, and is available free to ac
ademics\, researchers\, and non-commercial users. Our goal is to enable us
ers\, particularly non computer-scientists\, to treat a computer cluster a
s a forensic\, diagnostic\, or analytic tool. The talk will describe the d
etails of the system and the characteristics of some of the applications t
hat have been run on it.\n\nhttps://indico.tifr.res.in/indico/conferenceDi
splay.py?confId=435
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=435
END:VEVENT
BEGIN:VEVENT
SUMMARY:Kolmogorov Complexity
DTSTART;VALUE=DATE-TIME:20100129T090000Z
DTEND;VALUE=DATE-TIME:20100129T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-464@cern.ch
DESCRIPTION:I will introduce the basic definitions and framework of Kolmog
orov Complexity.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=464
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=464
END:VEVENT
BEGIN:VEVENT
SUMMARY:Encapsulating Quantification in the Typed Variables
DTSTART;VALUE=DATE-TIME:20100202T060000Z
DTEND;VALUE=DATE-TIME:20100202T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-439@cern.ch
DESCRIPTION:The asymmetry in the translation of Natural language sentences
involving existential and universal quantifiers is well known. It is poss
ible to get rid of this asymmetry by postulating 'quantified typed' variab
les. In this presentation\, we define a `quantified typed' variable and th
e algebra associated with these variables to prove the deductions using th
e method of reductio-ad-absurdum.\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=439
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=439
END:VEVENT
BEGIN:VEVENT
SUMMARY:Is this Function Linear?
DTSTART;VALUE=DATE-TIME:20100205T104500Z
DTEND;VALUE=DATE-TIME:20100205T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-480@cern.ch
DESCRIPTION:Given a function as black box\, where one can only query locat
ions in the truth table of the function and each query will be charged\, b
y querying only three locations one can tell whether the given function is
linear or not with high probability.\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=480
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=480
END:VEVENT
BEGIN:VEVENT
SUMMARY:Incorporating Views in Mathematical Models: An Approach Based on E
ntropy
DTSTART;VALUE=DATE-TIME:20100209T120000Z
DTEND;VALUE=DATE-TIME:20100209T130000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-426@cern.ch
DESCRIPTION:A mathematical model based on historical data or general past
experience may at times be an unsatisfactory model for the future. One way
to come up with a more accurate model is to explicitly incorporate in it
views that are believed to better reflect the future. We address this issu
e by letting $mu$ denote the original probability measure of a mathematica
l model. We then search for a probability measure $nu$ that minimizes a di
stance measure with respect to $mu$ and satisfies certain user specified v
iews or constraints. We consider Kullbach-Leibler distance as well as othe
r $f$-divergences as measures of distance between probability measures. We
show that under the KL distance\, our optimization problem may lack a clo
sed form solution when views involve fat tailed distributions. This drawba
ck may be corrected if a ``polynomial-divergence" is used. We also discuss
the optimal solution structure under these divergences when the views inc
lude those on marginal probabilities associated with the original probabil
ity measure. We apply these results to the area of portfolio optimization
where we note that a reasonable view that a particular portfolio of assets
has heavy tailed losses leads to a more realistic model through our appro
ach.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=426
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=426
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction to Parameterized Complexity
DTSTART;VALUE=DATE-TIME:20100212T104500Z
DTEND;VALUE=DATE-TIME:20100212T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-498@cern.ch
DESCRIPTION:For problems that are NP hard\, till now we have nothing bette
r than algorithms with exponential running times to obtain the optimal sol
ution.\n\nHowever\, if the time complexity of the algorithm has the form f
(k).p(n) where k and n are output and input sizes respectively\, p(n) is a
polynomial over n\, and f is a function depending only on k\, then the al
gorithm performs efficiently over small values of k.\n\nProblems having su
ch algorithms are said to be Fixed Parameter Tractable or in the class FPT
. We will study some fixed-parameter algorithms for certain common NP hard
problems.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=498
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=498
END:VEVENT
BEGIN:VEVENT
SUMMARY:Proportional Pricing Mechanisms for Utility Maximization in Shared
Computing Facilities
DTSTART;VALUE=DATE-TIME:20100216T103000Z
DTEND;VALUE=DATE-TIME:20100216T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-478@cern.ch
DESCRIPTION:Motivated by the recent advent of cloud computing facilities t
hat offer online computing power on demand\, we consider a large service f
acility that offers simultaneous service to a large number of heterogeneou
s jobs\, each associated with a distinct user. The execution time of each
job depends on the amount of resources applied to it. Our main concern her
e is in maximizing the social utility\, which comprises of the users' serv
ice utility minus their delay cost. This requires regulating the distribut
ion of available resources between the active users\, as well as controlli
ng the arrival rates of the different job types. In this work\, we formula
te a fluid queueing model that captures the essential ingredients of this
problem. We proceed to characterize the social optimum\, and propose a pri
cing mechanism that uniquely induces that social optimum\, without requiri
ng the system to be aware of the users' attributes and preferences. The pr
oposed mechanism divides the available resources between the active users
according to their price bids\, and is reminiscent of proportionally fair
pricing mechanisms of network flow control. We further consider briefly so
me relations with revenue-maximizing prices\, and stability properties of
the proposed price setting mechanism (joint work with Ishai Menache and As
u Ozduglar\, MIT).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=478
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=478
END:VEVENT
BEGIN:VEVENT
SUMMARY:Resource Sharing in Networks
DTSTART;VALUE=DATE-TIME:20100217T090000Z
DTEND;VALUE=DATE-TIME:20100217T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-451@cern.ch
DESCRIPTION:How should flows through a network be organized so that resour
ces are shared fairly\, and so that the network operates in a stable and e
fficient manner? This question arises in a number of application areas\, i
ncluding communication and transportation networks.\n\nThis talk will revi
ew definitions of fairness\, with particular emphasis on some of the attra
ctive features of proportional fairness. Next the talk will describe stoch
astic models of network routing and resource allocation\, when routing and
scheduling policies are designed to implement proportional fairness. Part
icular examples discussed will include Internet congestion control and ram
p metering policies for motorway networks.\n\nBio: Frank Kelly is Professo
r of the Mathematics of Systems\, University of Cambridge\, and Master of
Christ's College. His main research interests are in random processes\, ne
tworks and optimization. He is especially interested in applications to th
e design and control of networks and to the understanding of self-regulati
on in large-scale systems.\n\nFrank Kelly has received several prizes for
his work. In 1979 he won the Davidson Prize of the University of Cambridge
. In 1989 he was awarded the Guy Medal in Silver of the Royal Statistical
Society\, and in the same year he was elected a Fellow of the Royal Societ
y. He was awarded the 1991 Lanchester Prize of the Institute for Operation
s Research and the Management Sciences\, and in 1997 the Naylor Prize of t
he London Mathematical Society. In 2005 he received the IEEE Koji Kobayash
i Computers and Communications Award\, in 2008 the John von Neumann Theory
Prize\, and in 2009 the SIGMETRICS Achievement Award and the Gold Medal o
f the Association of European Operational Research Societies.\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=451
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=451
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stability Results for Stochastic Approximation Iterates
DTSTART;VALUE=DATE-TIME:20100219T104500Z
DTEND;VALUE=DATE-TIME:20100219T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-517@cern.ch
DESCRIPTION:In this talk\, after a brief introduction to stochastic approx
imation\, we shall present stability results for stochastic approximation
iterates. These are of the form: under certain assumptions the stochastic
approximation iterates remain bounded almost surely.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=517
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=517
END:VEVENT
BEGIN:VEVENT
SUMMARY:Laplacian Spectrum of Graphs\, Structural Distance and Evolutionar
y Relationship of Networks
DTSTART;VALUE=DATE-TIME:20100226T080000Z
DTEND;VALUE=DATE-TIME:20100226T090000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-510@cern.ch
DESCRIPTION:Existing graph measures can retrieve certain structural inform
ation\, but are not sufficient to capture all aspects of a network. We dev
elop a tool\, spectra of normalized graph Laplacian\, that helps to unders
tand the network structure with deep perception. Evolution of a network wi
th different graph operations produces specific eigenvalues. Thus\, useful
hypotheses about evolutionary processes can be made by investigating the
spectra of real networks. We also demonstrate that real networks of differ
ent classes pose characteristic eigenvalue distributions. Applying a meani
ngful distance measure\, we show that network structures are more similar
within the same class than between classes. We analyze 43 metabolic networ
ks from different species and find a separation of the three domains\, Bac
teria\, Archaea and Eukarya.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=510
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=510
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction to Parameterized Complexity
DTSTART;VALUE=DATE-TIME:20100226T104500Z
DTEND;VALUE=DATE-TIME:20100226T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-543@cern.ch
DESCRIPTION:For problems that are NP hard\, till now we have nothing bette
r than algorithms with exponential running times to obtain the optimal sol
ution.\n\nHowever\, if the time complexity of the algorithm has the form f
(k).p(n) where k and n are output and input sizes respectively\, p(n) is a
polynomial over n\, and f is a function depending only on k\, then the al
gorithm performs efficiently over small values of k.\n\nProblems having su
ch algorithms are said to be Fixed Parameter Tractable or in the class FPT
. We will study some fixed-parameter algorithms for certain common NP hard
problems (this is in continuation of the talk given on 12/02/2010).\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=543
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=543
END:VEVENT
BEGIN:VEVENT
SUMMARY:Effective Eilenberg Machines
DTSTART;VALUE=DATE-TIME:20100317T090000Z
DTEND;VALUE=DATE-TIME:20100317T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-555@cern.ch
DESCRIPTION:In this talk I revisit the idea of Eilenberg (from the book "A
utomata\, Languages and machines Vol.A" 1974) to use automata labelled wit
h relations as a general computational model. The model is now called Eile
nberg machines. Originally it was presented in an abstract mathematical wa
y. We make the model more precise to make it effective and give design cho
ices for simulation using modern programming languages. In contrast with T
uring machines\, it is interesting to mention that Eilenberg machines may
help to solve realistic problems\, like the Sanskrit sandhi inversion prob
lem for sentence segmentation as shown by Gérard Huet. We will also discu
ss how the simulation can be implemented in a clean way and proved correct
using a software of modern formal mathematics. We will also review the st
ate of the art of compilation of non-deterministic automata from regular e
xpressions since the latter shall be used as a language for describing the
control component of Eilenberg machines.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=555
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=555
END:VEVENT
BEGIN:VEVENT
SUMMARY:Contemporary Formal Mathematics (Part I)
DTSTART;VALUE=DATE-TIME:20100326T093000Z
DTEND;VALUE=DATE-TIME:20100326T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-572@cern.ch
DESCRIPTION:Arithmetic is at the dawn of any scientific discipline. In our
life we all have experienced (doing calculation) the validity of the foll
owing kind of equality 1947+1789 = 1789+1947 = 3736. It is less clear that
every scientist has ever proved this property of commutativity in the gen
eral case. Following the custom of the student seminar\, we will altogethe
r construct a very precise proof of this property. It will be conducted in
a context of contemporary formal mathematics\, in such a way that a stupi
d (but powerful) computer can check the proof. The consequence of this wor
k in our everyday life is enormous\, instead of saying "I believe that the
addition is commutative" one will now be proud to say "I know that the ad
dition is commutative".\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=572
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=572
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lower Bounds for Noisy Computations
DTSTART;VALUE=DATE-TIME:20100416T090000Z
DTEND;VALUE=DATE-TIME:20100416T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-605@cern.ch
DESCRIPTION:We consider noisy distributed computation in the setting of wi
reless sensor networks\, where processors are placed randomly on a unit sq
uare. They communicate with each other by broadcasting messages in order t
o compute some function whose input is distributed among them. Constraints
on power usage limit the range and the number of transmissions. Furthermo
re\, messages often get corrupted during transmission. Recently\, protocol
s have been proposed to compute efficiently in this model. We show the fir
st lower bounds in this model: in order to compute the PARITY or MAJORITY
of N bits reliably\, at least N log log N transmission are necessary. This
shows that the currently known protocols for these functions are optimal.
The techniques developed in this work can also be used to prove lower bou
nds for noisy decision trees.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=605
LOCATION:Colaba Campus D-405
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=605
END:VEVENT
BEGIN:VEVENT
SUMMARY:Information Theory in the Crowd
DTSTART;VALUE=DATE-TIME:20100420T103000Z
DTEND;VALUE=DATE-TIME:20100420T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-602@cern.ch
DESCRIPTION:In a short span in the 1940's Claude Shannon established the f
ields of information theory and modern cryptography which lie at the found
ation of the digital era. While the founder's legacy continues to influenc
e both fields\, the central challenges facing them today are very differen
t from the days of their origin. I will present recent work on both the fi
elds.\n\nThe point-to-point communication problem Shannon introduced in hi
s seminal paper founding the field of information theory is now thoroughly
understood and it has had a tremendous impact on the design of communicat
ion systems. This has provided the theoretical underpinnings for the wirel
ess communication revolution of the past two decades. What fundamentally l
imits the performance of all wireless communication networks today is inte
rference -- a necessarily network phenomenon. I will discuss a modern view
of interference: one which turns the root cause of interference\, namely
the broadcast nature of wireless transmissions\, into an ally by exploitin
g the potential for cooperation between nodes. I will present information
theoretically efficient communication schemes for canonical models of wire
less networks that cover a range of issues in wireless networks.\n\nIn lay
ing the foundations of modern cryptography\, Shannon considered the proble
m of secure communication where two parties want to communicate while excl
uding a third-party eavesdropper. But more sophisticated problems in crypt
ography today involve protecting the privacy of collaborators from each ot
her\, rather than from an external eavesdropper\, e.g.\, hospitals want to
pool their patient databases and perform data-mining without breaching pr
ivacy laws which restrict them from sharing sensitive information. Secure
multiparty computation is a theoretical framework for studying such proble
ms. I will present the best available upperbound on the efficiency of an i
mportant class of problems called secure two-party sampling. The bound is
based on a key tool we developed by generalizing the concept of common inf
ormation in information theory.\n\nBio: Vinod Prabhakaran received his Ph.
D. in December of 2007 from the EECS Department\, University of California
\, Berkeley. Since then he has been a postdoctoral researcher at Coordinat
ed Science Laboratory\, University of Illinois\, Urbana-Champaign. His res
earch interests are in information theory\, wireless communication\, crypt
ography\,\ndistributed signal processing and communication. He has receive
d a Commendation of the Senate of the Indian Institute of Science\, Bangal
ore\, the Tong Leong Lim Pre-Doctoral Prize and the Demetri Angelakos Memo
rial Achievement Award from the EECS Department\, University of California
\, Berkeley.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=602
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=602
END:VEVENT
BEGIN:VEVENT
SUMMARY:Probably Approximately CORRECT Learning
DTSTART;VALUE=DATE-TIME:20100423T090000Z
DTEND;VALUE=DATE-TIME:20100423T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-621@cern.ch
DESCRIPTION:We will see the definition of Probably Approximately CORRECT L
earning. Then we will prove that its easy to learn about Rectangles and Co
njunctions but hard to learn about 3-Term Disjunctions.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=621
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=621
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sparse Solutions to Under-determined Systems of Linear Equations
DTSTART;VALUE=DATE-TIME:20100430T090000Z
DTEND;VALUE=DATE-TIME:20100430T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-630@cern.ch
DESCRIPTION:An under-determined system of linear equation has infinitely m
any solutions (if it has a solution). But\, if we have some prior knowledg
e saying that the solution vector is "sparse" enough\, then under certain
conditions there exists a unique solution. Moreover we can find the soluti
on in polynomial time by solving a linear program. This is a celebrated re
sult (by Candes and Tao) that has created a whole new area called "*Compre
ssed sensing*". In this talk I will introduce the problem\, and will prove
a simple version of such reconstruction of sparse vectors from an under-d
etermined system of linear equations.\n\n(For everything that you want to
know about compressed sensing\, visit http://dsp.rice.edu/cs)\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=630
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=630
END:VEVENT
BEGIN:VEVENT
SUMMARY:Perturbed Identity Matrices have High Rank: Proof and some Applica
tions
DTSTART;VALUE=DATE-TIME:20100514T090000Z
DTEND;VALUE=DATE-TIME:20100514T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-638@cern.ch
DESCRIPTION:We will discuss a lower bound (due to Noga Alon) on the rank o
f any real matrix in which all the diagonal entries are significantly larg
er (in absolute value) than all the other entries. The fact that such matr
ices have "high" rank\, has helped in proving various results in combinato
rics. In this talk\, we would see some applications to - Geometry (we woul
d see that the bound obtained by the Johnson-Lindenstrauss lemma is almost
tight)\, and - Coding theory (we would see an upper bound on the size of
an Epsilon-balanced code).\n\n(See "Perturbed identity matrices have high
rank: proof and application" by Noga Alon for more details).\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=638
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=638
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Glimpse of Future Technologies Research at Intel
DTSTART;VALUE=DATE-TIME:20100517T053000Z
DTEND;VALUE=DATE-TIME:20100517T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-636@cern.ch
DESCRIPTION:We will present a broad range of longer-term research at Intel
Labs\, ranging from sensing/perception to networking to exascale computin
g. Much of this research is multi-disciplinary\, presenting rich new oppor
tunities for application of traditional analytical disciplines.\n\nBio: Su
meet Sandhu most recently served as Technical Advisor to the Vice Presiden
t of Research/Director of Future Technologies Research in Intel Labs. This
role provided exceptional exposure to diverse technologies such as sensin
g/perception\, cloud computing\, bio-systems\, energy\, optics/networking\
, ethnography\, robotics\, exascale computing\, and university-industry IP
policies. She developed the company-wide strategy for funding research in
Mobile Computing and Internet\, and has fostered edgy new projects in pro
babilistic computing\, energy storage\, and datacenter networking. These w
ere in addition to TA duties such as developing content for VP/CTO/Chairma
n talks\, representing the VP internally and externally\, visiting leading
technical institutions across US\, Europe\, India\, China\, Taiwan\, Midd
le East to develop deep perspective\, managing complex cross-organizationa
l efforts\, international visits and high level agendas\, and running a st
aff of senior research directors.\n\nPreviously\, she led the wireless 'Di
stributed Communication' research project on multi-hop/cooperative relays\
, network coding\, wideband MIMO and cellular scaling\, managing a team of
PhD scientists and engineers. The research won several best paper awards\
, was featured in top-100 IEEE papers in all disciplines in 2007\, and led
to the foundation of the new WiMax relay standard IEEE 802.16j. Sumeet ho
lds many patents reading on millions of 802.11n products worldwide (e.g. i
nvented the space-frequency interleaver\, and preambles for mixed networks
of single and multiple antenna devices). She has over 15 years of experie
nce in wireless: industry standards (WiFi\, WiMax\, PACS\, CDMA...)\, theo
retical research\, lab implementation\, field testing of cellular systems\
, and link/system level design. She has a PhD from Stanford University and
a BS and MS from the Massachusetts Institute of Technology\, all in Elect
rical Engineering. She held positions at Iospan Wireless\, Sprint Corporat
ion\, Hughes Research Laboratories and AT&T Bell Laboratories prior to Int
el.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=636
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=636
END:VEVENT
BEGIN:VEVENT
SUMMARY:Reducibility Among Fractional Stability Problems
DTSTART;VALUE=DATE-TIME:20100520T103000Z
DTEND;VALUE=DATE-TIME:20100520T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-639@cern.ch
DESCRIPTION:In a landmark paper\, Papadimitriou introduced several syntact
ic subclasses of the search class TFNP (Total Function Nondeterministic Po
lynomial) based on proof styles that (unlike TFNP) admit complete problems
. One particularly notable subclass of TFNP is PPAD (Polynomial Parity Arg
uments on Directed graphs)\, which includes a number of important stabilit
y problems\, including the celebrated problem of finding Nash equilibria i
n matrix form games. A recent series of results has culminated in showing
that finding Nash equilibria in multi-player games is complete for PPAD.\n
\nIn this work\, we establish the PPAD-completeness of a number of outstan
ding stability problems that arise in diverse applications\, including the
following: Fractional Stable Paths Problem (FSPP) -- Internet routing\; C
ore of Balanced Games -- Economics and Game theory\; Scarf's Lemma -- Comb
inatorics\; Hypergraph Matching - Social Choice and Preference Systems. In
fact\, we show that no fully polynomial-time approximation schemes exist
for these problems unless PPAD is in FP. In addition to enhancing our repe
rtoire of PPAD-complete problems\, we expect the concepts and techniques i
n this work to find future use in algorithmic game theory.\n\nIn the talk\
, I will motivate the fractional stability problems being studied and will
largely focus on the Fractional Stable Paths Problem. I will present two
new concepts that play a key role in our completeness reductions -- prefer
ence games and personalized equilibria. The talk will be largely self-cont
ained\; no significant knowledge of algorithmic game theory will be assume
d (joint work with Shiva Kintali\, Laura Poplawski\, Ravi Sundaram\, and S
hanghua Teng).\n\nBio: Rajmohan Rajaraman is a Professor of Computer Scien
ce at Northeastern University in Boston. He received his Bachelor's in Com
puter Science at IIT Kanpur in 1991 and his PhD in Computer Science at the
University of Texas at Austin in 1997. He was a postdoctoral fellow at DI
MACS before joining the Northeastern faculty in 1998. His research interes
ts include approximation algorithms\, algorithmic game theory\, distribute
d computing\, and complex networks.\n\nhttps://indico.tifr.res.in/indico/c
onferenceDisplay.py?confId=639
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=639
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Non-cooperative Perspective of Multi-user Communications
DTSTART;VALUE=DATE-TIME:20100521T053000Z
DTEND;VALUE=DATE-TIME:20100521T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-641@cern.ch
DESCRIPTION:The scarcity of spectrum is becoming an impediment to the grow
th of more capable wireless networks. Several measures are sought to addre
ss this problem: Freeing up unused spectrum\, sharing of spectrum through
new paradigms such as cognitive radio sensing\, as well as sophisticated c
ommunication schemes that rely on user cooperation.\n\nHowever\, given tha
t cooperation can lead to rate degradation when sharing spectrum\, this ma
y fail to happen with selfish users. This leads us to question: why would
users cooperate in sharing? What happens if they don’t? How can we incen
tivize them to cooperate? We consider these questions in a prototypical co
mmunication channel model\, the Gaussian interference channel in the conte
xt of cognitive radio systems. We propose contract designs for spectrum sh
aring using multi-user communication schemes that are robust to strategic
manipulation by participating players\, and can be made robust to the Mora
l hazard problem through simple enhancements. These can be seen as a stepp
ing stone towards development of a Non-cooperative framework for Multi-use
r Communication Theory.\n\nBio: Rahul Jain is an Assistant Professor in th
e EE and ISE departments in the Viterbi School of Engineering of the Unive
rsity of Southern California\, Los Angeles. He received his B.Tech in EE f
rom IIT Kanpur\, an MA in Statistics and a PhD in EECS from the University
of California\, Berkeley. He has diverse research interests with current
focus on Game Theory and Economics of Networks\, and Stochastic Control an
d Learning. He is a recipient of the Early Faculty Career Development (CAR
EER) Award from the National Science Foundation (NSF)\, and a James H. Zum
berge Faculty research and innovation award.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=641
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=641
END:VEVENT
BEGIN:VEVENT
SUMMARY:Communication in Subclasses of Petri Nets
DTSTART;VALUE=DATE-TIME:20100528T103000Z
DTEND;VALUE=DATE-TIME:20100528T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-648@cern.ch
DESCRIPTION:Petri net theory has nice theorems which relate subclasses of
Petri nets defined by structural conditions -- for example T-nets (also kn
own as marked graphs) and free choice nets -- to their behavioural propert
ies -- such as checking\, given a net together with an initial marking\, w
hether it exhibits a bounded amount of concurrency\, or whether a transiti
on can be fired\, or whether a transition can be fired from any reachable
marking. We argue that these theorems can be viewed as identifying nice co
mmunication patterns between different components of the given net (joint
work with Madhavan Mukund and Ramchandra Phawade).\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=648
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=648
END:VEVENT
BEGIN:VEVENT
SUMMARY:How to Generate a Large String almost Uniformly at Random\, with a
Small Number of Coin Tosses?
DTSTART;VALUE=DATE-TIME:20100528T084500Z
DTEND;VALUE=DATE-TIME:20100528T094500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-649@cern.ch
DESCRIPTION:Computer scientists devise randomized algorithms when they can
not find good deterministic ones. Then they try to decrease the randomness
used and still try to prove that the algorithm answers correctly with hig
h probability. They even hope to decrease the randomness used to a small a
mount that they can finally convert the randomized algorithm to a determin
istic one. A critical tool for doing this is to generate a long string fro
m a distribution close to uniform\, using only a small random string. We w
ill see how to do this using Epsilon Biased Probability Spaces and also co
nstruct more general objects called strings that are almost k-wise indepen
dent.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=649
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=649
END:VEVENT
BEGIN:VEVENT
SUMMARY:Streaming Algorithms using Finger Printing Technique
DTSTART;VALUE=DATE-TIME:20100604T090000Z
DTEND;VALUE=DATE-TIME:20100604T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-662@cern.ch
DESCRIPTION:Algorithms working on large data(say of size n) should be both
space and time efficient. Any algorithm has to see atleast the whole data
once\, so time required has to be atleast n(see sublinear time algorithms
for exceptions). Also it should use space much less than n(ie. o(n)). We
will see a good technique for designing such algorithms and use it to solv
e two problems. First will be checking whether a string x can be derived f
rom a grammar and the second one is to check whether a sequence of n numbe
rs are exactly the out degrees of the corresponding vertices in an n verte
x graph. Furthermore we will prove that the algorithms are optimal in thei
r space usage(ie. being more space efficient is a mathematical impossibili
ty) using techniques from communication complexity (joint work with Ajesh
Babu and Nutan Limaye and is going to be presented next Monday at TAMC 201
0).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=662
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=662
END:VEVENT
BEGIN:VEVENT
SUMMARY:Percolation
DTSTART;VALUE=DATE-TIME:20100621T050000Z
DTEND;VALUE=DATE-TIME:20100621T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-674@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
674
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=674
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Extentsions to the Theory of Regularity in Formal Languages
DTSTART;VALUE=DATE-TIME:20100625T053000Z
DTEND;VALUE=DATE-TIME:20100625T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-679@cern.ch
DESCRIPTION:In formal languguage theory\, regularity is a robust notion ha
ving many equivalent representations: regular expressions\, ﬁnite state
automata\, MSO etc. These have made important impact on programming and ve
riﬁcation tools. In this talk we delve into some the alternate formulati
ons and extensions of regularity:\n\n1) We will look at a new representati
on for the $epsilon$-free fragment of regular languages by replacing the c
onventional concatenation operator by the chop operator\, giving rise to t
he notion of chop expressions. We study the eﬀective conversions between
chop expressions\, regular expression and ﬁnite state automata and pin
down the complexities of various decision procedures for chop expressions
and its extensions. Following Salomaa’s complete axiomatization of regul
ar expressions\, we provide a complete axiom system for equality chop expr
essions.\n\n2) We shall survey the ﬁeld of automata over inﬁnite alpha
bet. This area of research aims at extending regular language theory over
ﬁnite alphabet to a corresponding theory over inﬁnite alphabet. We als
o provide an eﬃcient decision procedure for emptiness of nested data wor
d CMAs.\n\n3) We shall also discuss a logspace randomized 1-pass algorithm
for the membership checking of deterministic linear context-free language
s.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=679
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=679
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Near-tight Approximation Algorithm for the Robot Localization Pr
oblem
DTSTART;VALUE=DATE-TIME:20100628T060000Z
DTEND;VALUE=DATE-TIME:20100628T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-676@cern.ch
DESCRIPTION:Localization is a fundamental problem in robotics. The "kidnap
ped robot" possesses a compass and map of its environment\; it must determ
ine its location at a minimum cost of travel distance. The problem is NP-h
ard even to minimize within factor $clog n$\, where $n$ is the map size. N
o efficient approximation algorithm is known.\n\nWe give an $O(log3 n)$-fa
ctor algorithm which runs in polynomial time. The key idea is to plan trav
el in a ``majority-rule'' map\, which eliminates uncertainty and permits a
link to the $frac{1}{2}$-Group Steiner (not Group Steiner) problem. The a
pproximation factor is not far from optimal: we prove a $clog^{2-epsilon}
n$ lower bound\, assuming $NP notsubseteq ZTIME(n^{polylog(n)})$\, for the
grid graphs commonly used in practice. We also extend the algorithm to po
lygonal maps by discretizing the problem using novel geometric techniques.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=676
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=676
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hats Off to Theoretical Computer Science
DTSTART;VALUE=DATE-TIME:20100628T103000Z
DTEND;VALUE=DATE-TIME:20100628T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-686@cern.ch
DESCRIPTION:The speaker will discuss three puzzles in this talk and using
ideas from theoretical computer science try to find answers for them.\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=686
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=686
END:VEVENT
BEGIN:VEVENT
SUMMARY:Catalysis in Mass Action Kinetics
DTSTART;VALUE=DATE-TIME:20100702T090000Z
DTEND;VALUE=DATE-TIME:20100702T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-693@cern.ch
DESCRIPTION:If networks of chemical reactions are the circuits of biology
then catalysts are the transistors\, or perhaps switches. But which specie
s should be called catalysts? Come to the talk and find out. In the real w
orld\, "transistors" have "leakage currents" which make the circuit design
er's life harder. In contrast\, it turns out that leakage currents make an
alysis simpler. Some elementary algebraic geometry shows that leakage curr
ents rule out "relevant siphons\," and make the dynamics particularly well
-behaved. This will be a chalk talk. All are welcome but the talks are aim
ed at those comfortable with theorems and proofs.\n\nhttps://indico.tifr.r
es.in/indico/conferenceDisplay.py?confId=693
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=693
END:VEVENT
BEGIN:VEVENT
SUMMARY:Complexity of Graphical Realizations of Linear Codes
DTSTART;VALUE=DATE-TIME:20100705T053000Z
DTEND;VALUE=DATE-TIME:20100705T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-650@cern.ch
DESCRIPTION:A graphical realization of a linear code C consists of an assi
gnment of the coordinates of C to the vertices of a graph\, along with a s
pecification of linear state spaces and linear "local constraint" codes to
be associated with the edges and vertices\, respectively\, of the graph.
Special cases of such realizations are trellises\, Tanner graphs\, and ind
eed any factor graph. A graphical realization of C specifies an associated
sum-product decoding algorithm for C. The computational complexity of the
associated sum-product algorithm is largely determined by the dimensions
of the local constraint codes in the realization. Thus\, the complexity of
a graphical realization may be defined in terms of the dimensions of the
local constraint codes. In this talk\, we will give an overview of graphic
al realizations\, and what is known about the complexity of graphical real
izations of a linear code.\n\nBio: Navin Kashyap received the B.Tech. degr
ee in Electrical Engineering from the Indian Institute of Technology\, Bom
bay\, in 1995\, the M.S. degree in Electrical Engineering from the Univers
ity of Missouri-Rolla in 1997\, and the M.S. degree in Mathematics and the
Ph.D. degree in Electrical Engineering from the University of Michigan\,
Ann Arbor\, in 2001.\n\nFrom November 2001 to November 2003\, he was a pos
tdoctoral research associate at the University of California\, San Diego.
In January 2004\, he joined the Department of Mathematics and Statistics a
t Queen's University\, Kingston\, Ontario\, where he is currently an Assoc
iate Professor. His research interests lie primarily in the application of
combinatorial methods in information and coding theory.\n\nDr. Kashyap is
currently visiting the ECE Department at the Indian Institute of Science\
, Bangalore\, on sabbatical from Queen's.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=650
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=650
END:VEVENT
BEGIN:VEVENT
SUMMARY:Counting paths in VPA is complete for #NC1
DTSTART;VALUE=DATE-TIME:20100708T090000Z
DTEND;VALUE=DATE-TIME:20100708T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-696@cern.ch
DESCRIPTION:We prove a $#$NC$^1$ upper bound for the problem of counting a
ccepting paths in any fixed visibly pushdown automaton. Our algorithm invo
lves a non-trivial adaptation of the arithmetic formula evaluation algorit
hm of Buss\, Cook\, Gupta\, Ramachandran. We also show that the problem is
$#$NC$^1$ hard. Our results show that the difference between $#$BWBP and
$#$NC$^1$ is captured exactly by the addition of a visible stack to a nond
eterministic finite-state automata.\n\nTo appear in proceedings of the 16t
h International Computing and Combinatorics Conference COCOON\, 19-21 July
\, 2010\, Vietnam. (Andreas Krebs (University of Tubingen\, Germany)\, Nut
an Limaye (TIFR\, Mumbai)\, Meena Mahajan (Institute of mathematical scien
ces\, Chennai)\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=696
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=696
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Near Optimal Algorithm for Finding Euclidean Shortest Path in Po
lygonal Domain
DTSTART;VALUE=DATE-TIME:20100713T060000Z
DTEND;VALUE=DATE-TIME:20100713T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-701@cern.ch
DESCRIPTION:The Euclidean shortest path problem in a polygonal region is o
ne of the oldest and best-known in Computational Geometry due to its vario
us applications. Given a polygon with holes $P$ and two points $s$ and $t$
interior to it\, our algorithm finds an Euclidean shortest path from $s$
to $t$ in $O(n+m(lg{m})(lg{n}))$ time using $O(n)$ space. Here\, $n$ is th
e number of vertices in the given polygonal domain\, and $m$ is the number
of holes. This problem is listed as part of The Open Problems Project\, w
hich intends for a solutions with $O(n+m(lg{m}))$ time using $O(n)$ space.
After identifying hourglasses in $P$\, the regions of interest in $P$ are
traversed with the shortest path wavefront.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=701
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=701
END:VEVENT
BEGIN:VEVENT
SUMMARY:QIP = PSPACE
DTSTART;VALUE=DATE-TIME:20100714T090000Z
DTEND;VALUE=DATE-TIME:20100714T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-707@cern.ch
DESCRIPTION:We prove that the complexity class QIP\, which consists of all
problems having quantum interactive proof systems\, is contained in PSPAC
E. This containment is proved by applying a parallelized form of the matri
x multiplicative weights update method to a class of semidefinite programs
that captures the computational power of quantum interactive proofs. As t
he containment of PSPACE in QIP follows immediately from the well-known eq
uality IP = PSPACE\, the equality QIP = PSPACE follows (joint work with Zh
engfeng Ji\, Sarvagya Upadhyay and John Watrous).\n\nWinner of the Best Pa
per Award STOC 2010.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=707
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=707
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimal Homologous Cycles\, Total Unimodularity\, and Linear Progr
amming
DTSTART;VALUE=DATE-TIME:20100715T053000Z
DTEND;VALUE=DATE-TIME:20100715T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-697@cern.ch
DESCRIPTION:Given a simplicial complex with weights on its simplices\, and
a nontrivial cycle on it\, we are interested in finding the cycle with mi
nimal weight which is homologous to the given one. A recent result showed
that this problem is NP-hard when the homology is defined using binary coe
fficients\, which is intuitive and easy to deal with. In this paper we con
sider homology defined with integer coefficients. We show that the boundar
y matrix of a finite simplicial complex is totally unimodular if and only
if the simplicial complex is relatively torsion-free with the homology def
ined relative to all pure subcomplexes of appropriate dimensions. Because
of the total unimodularity of the boundary matrix\, we can solve the optim
ization problem\, which is inherently an integer programming problem\, as
a linear program and obtain an integer solution. Thus the problem of findi
ng optimal cycles in a given homology class can be solved in polynomial ti
me. One consequence of our result\, among others\, is that one can compute
in polynomial time an optimal (d-1)-cycle in a given homology class for a
ny triangulation of an orientable compact d-manifold or for any finite sim
plicial complex embedded in d-dimensional space. Our optimization approach
can also be used for various related problems\, such as finding an optima
l chain to a given one when these are not cycles. Our result can also be v
iewed as providing a topological characterization of total unimodularity (
this is joint work with Tamal Dey at Ohio State University and Anil Hirani
at University of Illinois. The paper was accepted to STOC'10\, and is ava
ilable on arXiv: http://arxiv.org/abs/1001.0338.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=697
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=697
END:VEVENT
BEGIN:VEVENT
SUMMARY:Towards a Portable Execution Model for Extreme Scale Multicore Sys
tems
DTSTART;VALUE=DATE-TIME:20100721T060000Z
DTEND;VALUE=DATE-TIME:20100721T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-698@cern.ch
DESCRIPTION:The computer industry is at a major inflection point in its ha
rdware roadmap due to the end of a decades-long trend of exponentially inc
reasing clock frequencies. Computer systems anticipated in the 2015 -- 202
0 timeframe are referred to as Extreme Scale because they will be built us
ing homogeneous and heterogeneous many-core processors with 100's of cores
per chip. These systems pose new critical challenges for software in the
areas of concurrency\, energy efficiency and resiliency. Unlike previous g
enerations of hardware evolution\, this shift towards many-core computing
will have a profound impact on software. These software challenges are fur
ther compounded by the need to enable parallelism in workloads and applica
tion domains that have traditionally not had to worry about multiprocessor
parallelism in the past.\n\nAn execution model serves a valuable role in
providing a shared conceptual view for all stakeholders in a computing pla
tform ecosystem. Successful execution models from the past (e.g.\, Von Neu
mann\, vector parallelism\, SMP parallelism\, Bulk-Synchronous parallelism
) were built on primitives that were well matched with past device\, archi
tecture\, and software technology trends\, but are mismatched to future mu
lticore systems where performance has to be driven by parallelism and cons
trained by energy. In this talk\, we identify key primitives that we belie
ve will be necessary for a successful execution model for future extreme s
cale systems. We will discuss the portability of these execution model pri
mitives based on their ability to support multiple programming models and
their amenability to be mapped to a wide range of extreme scale hardware.\
n\nWe present early experiences with these execution model primitives in t
he Habanero Multicore Software Research project at Rice University which t
argets mainstream homogeneous and heterogeneous multicore systems\, and di
scuss future directions in the context of the NSF Expeditions project on t
he Center for Domain-Specific Computing (http://www.cdsc.ucla.edu/) which
targets embedded systems with an initial focus on the medical imaging doma
in. Both projects takes a two-level approach to programming models\, with
a higher-level macro-dataflow model based on Intel Concurrent Collections
(CnC) for parallelism-oblivious domain experts \, and a lower-level task-p
arallel model based on the high productivity X10 language for parallelism-
aware developers. We discuss language\, compiler and runtime implementatio
n challenges that must be overcome to efficiently support these primitives
on future mainstream and embedded multicore systems.\n\nBio: Vivek Sarkar
conducts research in multiple aspects of parallel software including prog
ramming languages\, program analysis\, compiler optimizations and runtimes
for parallel and high performance computer systems. He currently leads th
e Habanero Multicore Software Research project at Rice University\, serves
as Associate Director of the NSF Expeditions project on the Center for Do
main-Specific Computing\, and as co-PI on the DARPA-funded project on Plat
form-Aware Compilation Environment (PACE). Prior to joining Rice in July 2
007\, Vivek was Senior Manager of Programming Technologies at IBM Research
. His responsibilities at IBM included leading IBM's research efforts in p
rogramming model\, tools\, and productivity in the PERCS project during 20
02- 2007 as part of the DARPA High Productivity Computing System program.
His past projects include the X10 programming language\, the Jikes Researc
h Virtual Machine for the Java language\, the ASTI optimizer used in IBM's
XL Fortran product compilers\, the PTRAN automatic parallelization system
\, and profile-directed partitioning and scheduling of Sisal programs. Viv
ek became a member of the IBM Academy of Technology in 1995\, the E.D. But
cher Professor in Engineering at Rice University in 2007\, and was inducte
d as an ACM Fellow in 2008. He holds a B.Tech. degree from the Indian Inst
itute of Technology\, Kanpur\, an M.S. degree from University of Wisconsin
-Madison\, and a Ph.D. from Stanford University. In 1997\, he was on sabba
tical as a visiting associate professor at MIT\, where he was a founding m
ember of the MIT RAW multicore project.\n\nLINKS:\n\n[1] Habanero Multicor
e Software Research project (http://habanero.rice.edu)\n[2] Habanero Concu
rrent Collections download (http://habanero.rice.edu/cnc)\n[3] Habanero Ja
va download (http://habanero.rice.edu/hj)\n[4] Overview article on "Softwa
re Challenges at Extreme Scale"\,\nhttp://www.scidacreview.org/1001/html/s
oftware.html.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=698
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=698
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Weighted Geometric Set Cover
DTSTART;VALUE=DATE-TIME:20100721T090000Z
DTEND;VALUE=DATE-TIME:20100721T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-716@cern.ch
DESCRIPTION:There has been much progress on geometric set cover problems\,
but most known techniques only apply to the unweighted setting. For the w
eighted setting\, very few results are known with approximation guarantees
better than that for the combinatorial set cover problem. In this talk\,
we employ the idea of quasi-uniform sampling to obtain improved approximat
ion guarantees in the weighted setting for a large class of problems for w
hich such guarantees were known in the unweighted case.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=716
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=716
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Universal and Differentially Private Steiner Trees and TSPs
DTSTART;VALUE=DATE-TIME:20100722T053000Z
DTEND;VALUE=DATE-TIME:20100722T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-705@cern.ch
DESCRIPTION:Given a collection of vertices in a network\, the problem of f
inding the minimum cost sub-network connecting a set of client nodes to a
dedicated hub-node is the Steiner tree problem. Consider the following two
twists. In the first setting\, the set of client nodes is unknown\, and t
he algorithm needs to output a master tree which "works well" for all plau
sible set of clients. In the second setting\, the set of clients is known\
, however\, the algorithm's output must respect privacy constraints and no
t reveal any information about a node (of it being a client or a not). The
se two settings\, defined similarly for the traveling salesman problem\, l
ead to the universal and private versions of the problem\, respectively. C
learly\, these need the design of new classes of algorithms and the questi
on that naturally arises is: how good/bad is the performance of these comp
ared to the traditional algorithms?\n\nAfter giving precise definitions of
the problems in hand\, we'll resolve the above question for the above two
problems by giving lower bounds which match the known upper bounds (which
also we'll sketch) up to constant factors. Our constructions are simple\,
and I'll attempt to convey them in this self contained talk (joint work w
ith Anand Bhalgat and Sanjeev Khanna\, both at University of Pennsylvania)
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=705
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=705
END:VEVENT
BEGIN:VEVENT
SUMMARY:Malware Detection by Behavioural Approach and Protection by Access
Control
DTSTART;VALUE=DATE-TIME:20100727T090000Z
DTEND;VALUE=DATE-TIME:20100727T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-714@cern.ch
DESCRIPTION:As computers and networked systems have become an integral par
t of our daily lives\, securing information from unauthorized access\, mis
use and modification has become very important. Malware\, short for malici
ous software\, is software designed to infiltrate a computer system withou
t the owner's informed consent and cause damage. The following factors cal
l for an immediate attention to the problem of securing information system
s from malware attacks\n\n1. Malware industry has become more organized\,
resulting in a rapid increase in the amount and the type of malware.\n2. T
here is a huge loss of revenue due to malware attacks.\n3. Solutions curre
ntly provided by the commercial anti-virus/security products are easily de
feated.\n4. The time gap between a malware outbreak and the malware carryi
ng out its intended damage is very small.\n\nWe have developed a new appro
ach for detecting malware by their behavioural characteristics\, and limit
ing the damage due to them by access control mechanisms. A brief overview
of the results obtained will be presented in this talk.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=714
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=714
END:VEVENT
BEGIN:VEVENT
SUMMARY:Local Search for Geometric Set Cover Problems
DTSTART;VALUE=DATE-TIME:20100728T053000Z
DTEND;VALUE=DATE-TIME:20100728T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-725@cern.ch
DESCRIPTION:In SocG 2009\, Mustafa and Ray showed that a local search algo
rithm is\, somewhat surprisingly\, a PTAS for certain geometric set cover/
hitting set problems. In this context\, a PTAS is a scheme that for any ep
s > 0\, yields an algorithm that runs in polynomial time and returns a sol
ution whose size is at most (1 + eps) times that of the optimal set cover/
hitting set. A similar PTAS was presented by Chan and Har-Peled at the sam
e conference for the largest independent set problem in certain geometric
intersection graphs.\n\nWe will discuss a hitting set result of Mustafa an
d Ray and\, time permitting\, an application of their machinery to a terra
in guarding problem by Gibson\, Kanade\, Krohn\, and the speaker.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=725
LOCATION:Colaba Campus D-405
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=725
END:VEVENT
BEGIN:VEVENT
SUMMARY:Persistence of Long-Range-Dependence Under Data Compression
DTSTART;VALUE=DATE-TIME:20100803T060000Z
DTEND;VALUE=DATE-TIME:20100803T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-728@cern.ch
DESCRIPTION:One of the early motivations for current interest in the stoch
astic networks community in the study of network models involving long-ran
ge-dependent stochastic processes was the observation\, based on statistic
al analysis of data\, that variable-bit-rate video traffic over networks a
ppears to exhibit long-range-dependent behavior. Such traffic is typically
placed on the network after data compression algorithms are used on an un
derlying video source. It is natural to ask what role the data compression
algorithm plays in the resulting long-range-dependent nature of the traff
ic.\n\nMotivated by this question we study the entropy density of an under
lying long-range-dependent process as a stochastic process in its own righ
t\, focusing on discrete time models. For classes of processes including r
enewal processes we prove that long-range-dependence of the underlying pro
cess implies long-range-dependence of the entropy density process\, with t
he same Hurst exponent.\n\nThe underlying background in the data compressi
on of stochastic processes\, including the fundamental lemma of Barron rel
ating the entropy density to data compression\, and existing results for t
he short-range-dependent case that have the same flavor as our results\, s
uch as those due to Kontoyiannis\, will also be discussed in this talk (jo
int work with Barlas Ou{g}uz.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=728
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=728
END:VEVENT
BEGIN:VEVENT
SUMMARY:Tail-robust Scheduling via Limited Processor Sharing
DTSTART;VALUE=DATE-TIME:20100805T060000Z
DTEND;VALUE=DATE-TIME:20100805T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-726@cern.ch
DESCRIPTION:From a rare events perspective\, scheduling disciplines that w
ork well under light (exponential) tailed workload distributions do not pe
rform well under heavy (power) tailed workload distributions\, and vice-ve
rsa\, leading to fundamental problems in designing schedulers that are rob
ust to distributional assumptions on the job sizes. We show how to exploit
partial workload information (system load) to design a scheduler that pro
vides robust performance across heavy-tailed and light-tailed workloads. S
pecifically\, we derive new asymptotics for the tail of the stationary soj
ourn time under Limited Processor Sharing (LPS) scheduling for both heavy-
tailed and light-tailed job size distributions\, and show that LPS can be
robust to the tail of the job size distribution if the multiprogramming le
vel is chosen carefully as a function of the load.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=726
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=726
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Stability and Sample Complexity of Stochastic Approximation Ite
rates
DTSTART;VALUE=DATE-TIME:20100831T090000Z
DTEND;VALUE=DATE-TIME:20100831T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-762@cern.ch
DESCRIPTION:(i) We propose and analyze a new scheme for stabilizing the st
ochastic approximation iterates\, viz.\, an adaptation of step sizes that
controls the growth of the iterates without affecting their asymptotic beh
avior. This amounts to scaling the step sizes appropriately when the itera
tes are sufficiently far away from the origin. Only a finite random number
of steps differ from the original scheme\, and unlike the projection meth
od\, this scheme does not introduce spurious equilibria. Further\, instead
of requiring the o.d.e. to descend the Lyapunov function everywhere where
the function isn't at its minimum\, we only require it to do so outside a
sphere of arbitrarily large radius. [http://arxiv.org/abs/1007.4689]\n\n(
ii) Sample complexity refers to the number of samples needed to be within
a specific distance from the attracting set of the o.d.e. with a given pro
bability provided that the iterates enter its domain of attraction with th
e step size sufficiently small at the time of entry. In previous work by B
orkar\, sample complexity estimates were obtained under the assumption tha
t the suitably re-scaled martingale difference noise remains bounded. We r
ecover these results under a much weaker condition that only requires the
re-scaled martingale differences to have an exponentially decaying conditi
onal tail probability. [http://arxiv.org/abs/1007.4684]\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=762
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=762
END:VEVENT
BEGIN:VEVENT
SUMMARY:Data Structures for Storing Small Sets in the Bitprobe Model
DTSTART;VALUE=DATE-TIME:20100903T113000Z
DTEND;VALUE=DATE-TIME:20100903T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-785@cern.ch
DESCRIPTION:We study the following set membership problem in the bitprobe
model: Given a set S from a finite universe U \, represent it in memory so
that membership queries of the form "Is x in S?" can be answered with a s
mall number of bitprobes. We obtain explicit schemes that come close to th
e information theoretic lower bound of Buhrman et al. [STOC 2000\, SICOMP
2002] and improve the results of Radhakrishnan et al. [ESA 2001] when the
size of sets and the number of probes is small.\n\nWe show that any scheme
that stores sets of size two from a universe of size m and answers member
ship queries using two bitprobes requires space Ω(m^(4/7)). The previous
best lower bound was Ω(m^(1/2}). This is the first instance where the inf
ormation theoretic lower bound is found to be not tight for adaptive schem
es.\n\nWe show that any non-adaptive three probe scheme for storing sets o
f size two from a universe of size m requires Ω(m^(1/2}) bits of memory.
This extends a result of Alon and Feige [SODA 2009] to small sets (to be p
resented in ESA 2010\; joint work with Jaikumar Radhakrishnan and Smit Sha
h).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=785
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=785
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximate Counting\, Uniform Generation and Rapidly Mixing Marko
v Chains
DTSTART;VALUE=DATE-TIME:20100906T083000Z
DTEND;VALUE=DATE-TIME:20100906T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-781@cern.ch
DESCRIPTION:Most algorithmic problems could be viewed as questions being a
sked about a natural binary relation and we can classify them into decisio
n\, search\, uniform generation and counting. For some relations\, countin
g and uniform generation could be NP-Hard even if decision or search could
be done efficiently. M. Jerrum and A. Sinclair used the Markov chain Mont
e Carlo method to approximately count the number of perfect matchings of a
dense bipartite graph. They reduced approximate counting to almost unifor
m generation of perfect matchings. Almost uniform generation was done by s
tarting with an arbitrary solution\, and repeatedly making random modifica
tions to it. This process was modeled as a Markov chain\, with a stationar
y distribution that is uniform on the set of perfect matchings. The rapid
convergence of the chain towards its stationary distribution\, was proved
by bounding a quantity called conductance. This was done using a novel met
hod of defining canonical paths between every pair of states in the chain.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=781
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=781
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Algebraic Properties of a Subclass of Finite Normal Form Game
s
DTSTART;VALUE=DATE-TIME:20100907T103000Z
DTEND;VALUE=DATE-TIME:20100907T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-788@cern.ch
DESCRIPTION:My talk will introduce the problem of computing all Nash equil
ibria of finite normal form games. I will define a subclass of finite norm
al form games and present:\n\n- a method for computing all its Nash equili
bria and\n- a method for deciding membership to the subclass.\n\nI will co
nclude my talk by presenting some properties of these games\, and an examp
le to show working of each of the presented methods.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=788
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=788
END:VEVENT
BEGIN:VEVENT
SUMMARY:Game Theory
DTSTART;VALUE=DATE-TIME:20100913T053000Z
DTEND;VALUE=DATE-TIME:20100913T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-757@cern.ch
DESCRIPTION:The course will cover Zero sum and Nonzero sum games\, Extensi
ve Form and Normal Form games\, Nash equilibrium\, Correlated equailibrium
\, Refinements\, Bayesian games\, Repeated Games\, Differential Games. Tim
e permitting Stochastic games will also be covered.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=757
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=757
END:VEVENT
BEGIN:VEVENT
SUMMARY:Game Theory
DTSTART;VALUE=DATE-TIME:20100914T053000Z
DTEND;VALUE=DATE-TIME:20100914T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-758@cern.ch
DESCRIPTION:The course will cover Zero sum and Nonzero sum games\, Extensi
ve Form and Normal Form games\, Nash equilibrium\, Correlated equailibrium
\, Refinements\, Bayesian games\, Repeated Games\, Differential Games. Tim
e permitting Stochastic games will also be covered.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=758
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=758
END:VEVENT
BEGIN:VEVENT
SUMMARY:Game Theory
DTSTART;VALUE=DATE-TIME:20100915T053000Z
DTEND;VALUE=DATE-TIME:20100915T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-759@cern.ch
DESCRIPTION:The course will cover Zero sum and Nonzero sum games\, Extensi
ve Form and Normal Form games\, Nash equilibrium\, Correlated equailibrium
\, Refinements\, Bayesian games\, Repeated Games\, Differential Games. Tim
e permitting Stochastic games will also be covered.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=759
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=759
END:VEVENT
BEGIN:VEVENT
SUMMARY:Let Us Catch Some Bugs
DTSTART;VALUE=DATE-TIME:20100917T113000Z
DTEND;VALUE=DATE-TIME:20100917T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-799@cern.ch
DESCRIPTION:In this talk\, we will discuss current research efforts to ide
ntify metamorphic viruses.\n\nhttps://indico.tifr.res.in/indico/conference
Display.py?confId=799
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=799
END:VEVENT
BEGIN:VEVENT
SUMMARY:Communicating Under Channel Uncertainty
DTSTART;VALUE=DATE-TIME:20101001T113000Z
DTEND;VALUE=DATE-TIME:20101001T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-842@cern.ch
DESCRIPTION:For a single transmit and receive antenna system\, a new const
ellation design is proposed to combat errors in the phase estimate of the
channel coefficient. The proposed constellation is a combination of PSK an
d PAM constellations\, where PSK is used to provide protection against pha
se error\, while PAM is used to increase the transmission rate using the k
nowledge of the magnitude of the channel coefficient. The performance of t
he proposed constellation is shown to be significantly better than the wid
ely used QAM in terms of probability of error. The proposed strategy can a
lso be extended to systems using multiple transmit and receive antennas.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=842
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=842
END:VEVENT
BEGIN:VEVENT
SUMMARY:Malware Analysis - 1
DTSTART;VALUE=DATE-TIME:20101008T113000Z
DTEND;VALUE=DATE-TIME:20101008T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-855@cern.ch
DESCRIPTION:In this talk\, we present the experimental methods and introdu
ce the tools for analyzing advanced malware.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=855
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=855
END:VEVENT
BEGIN:VEVENT
SUMMARY:Malware Analysis - 2
DTSTART;VALUE=DATE-TIME:20101015T113000Z
DTEND;VALUE=DATE-TIME:20101015T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-866@cern.ch
DESCRIPTION:In this talk\, we present the analysis of well-known email-wor
ms Netsky\, Beagle and MyDoom.\n\nhttps://indico.tifr.res.in/indico/confer
enceDisplay.py?confId=866
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=866
END:VEVENT
BEGIN:VEVENT
SUMMARY:Analyzing Bio-molecules Using Graphs\, Geometric Invariants and Ma
chine Learning
DTSTART;VALUE=DATE-TIME:20101102T103000Z
DTEND;VALUE=DATE-TIME:20101102T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-910@cern.ch
DESCRIPTION:The focus of the talk will be on analysis of protein structure
s using geometric\, machine learning and graph theoretic techniques and it
s implications in functional site predictions. Proteins are versatile bio-
molecules made up of amino acids and are involved in many cellular functio
ns. They function through an arrangement\, known as functional site\, of a
small number of spatially proximal amino acid residues (typically three t
o six) in the structure. Given a new protein\, biologists are interested i
n determining its functional site using suitable experimental techniques.
In doing so\, they are faced with enormous number of choices due to combin
atorial explosion and it is just not feasible to evaluate these choices ex
haustively because of excessive time and resource requirements. To overcom
e the problem\, we developed a method that provides the biologists with a
small list of most likely functional sites\, which will serve as a useful
guide while designing their experiments.\n\nIn our scheme\, we represent e
ach protein structure as an unweighted undirected graph with amino acid re
sidues being the nodes. The nodes are connected with an edge if the corres
ponding amino acid residues are spatially proximal in the structure. Each
functional site is represented as a clique in this set up. We extract cand
idate functional sites from each protein using Bron-Kerbosch clique findin
g algorithm. Now\, the objective is to determine most likely functional si
tes from a large number of candidate sites. The work is founded on the wel
l characterized biological knowledge that the functionally important subst
ructures are conserved and recur in functionally related proteins. We repr
esent the candidate functional sites using geometric invariants\, which re
main unchanged upon transformations like rotation and translation. The can
didate sites are grouped\, using machine learning techniques\, based on th
eir similarity in a space spanned by geometric invariants. The recurring c
andidate sites are analyzed to provide a rank list of possible functional
sites to the experimental biologists. Finally\, I will present a few examp
les of successful application of this method in novel proteins.\n\nhttps:/
/indico.tifr.res.in/indico/conferenceDisplay.py?confId=910
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=910
END:VEVENT
BEGIN:VEVENT
SUMMARY:Modem Design for MultiGigabit OFDM System
DTSTART;VALUE=DATE-TIME:20101115T083000Z
DTEND;VALUE=DATE-TIME:20101115T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-921@cern.ch
DESCRIPTION:In ``multiGigabit" systems having digital signal processing (D
SP) based transceiver architectures\, the analog-to-digital converter (ADC
) is a critical component due to the cost and power requirements of high p
recision ADC. Thus\, we may be forced to reduce the precision to 1-3 bits.
We study the performance degradation of a conventional OFDM receiver at l
ow precision and investigate an alternate receiver structure\, which we ca
ll analog OFDM receiver (AOR)\, where we incorporate some analog processin
g before sampling. In an ideal implementation\, the AOR yields near full p
recision performance. We also discuss some challenges in implementing such
a receiver.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=921
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=921
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Parallel Repetition Theorem and Related Results
DTSTART;VALUE=DATE-TIME:20101124T060000Z
DTEND;VALUE=DATE-TIME:20101124T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-986@cern.ch
DESCRIPTION:In a 2-Prover 1-Round Game\, a verifier draws a pair of questi
ons (X\,Y ) from a distribution D and sends one each to two co-operating\,
non-communicating players who need to respond back with answers A\,B. The
verifier checks the answers using a known predicate V(X\,Y\,A\,B)\, and d
eclares a win or loss. The aim of the players is to plan a strategy to win
the game with the highest probability over the set of questions. The n -f
old parallel repetition of the game has the verifier drawing n pairs of \,
i.i.d.\, and sending each player an n-tuple of questions. The players hav
e to respond with n answers each\, one for each co-ordinate. The Parallel
Repetition theorem proven by Raz states that the probability of winning th
e repeated game in all co-ordinates drops exponentially with n. The theore
m was a key result used in proving various hardness of approximation resul
ts. In this talk\, we give a brief overview of the implications of this th
eorem\, and various related results.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=986
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=986
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stochastic Approximation: Some Theory and an Application
DTSTART;VALUE=DATE-TIME:20101129T083000Z
DTEND;VALUE=DATE-TIME:20101129T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-993@cern.ch
DESCRIPTION:We discuss a vector minmax problem for controlled Markov chain
s [3]. The problem of controlling a finite state Markov chain in the prese
nce of an adversary so as to ensure desired performance levels for a vecto
r of objectives is cast in the framework of Blackwell approachability. Rel
ying on an elementary two time scale construction a control scheme is prop
osed which ensures almost sure convergence to the desired set regardless o
f the adversarial actions. This problem serves as an example of stochastic
approximation. We conclude with some general theoretical results in stoch
astic approximation. These relate to stability [2] and sample complexity [
1].\n\nReferences\n\n[1] S. Kamal\, On the convergence\, lock-in probabili
ty\, and sample complexity of stochastic approximation\, SIAM Journal on C
ontrol and Optimization\, Volume 48\, Number 8\, October 2010\, pp. 5178-5
192.\n[2] S. Kamal\, Stabilization of stochastic approximation by step siz
e adaptation\, Preprint available at http://arxiv.org/abs/1007.4689\n[3] S
. Kamal\, A vector minmax problem for controlled Markov chains\, Preprint
available at http://arxiv.org/abs/1011.0675\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=993
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=993
END:VEVENT
BEGIN:VEVENT
SUMMARY:Ryan Williams' New Results on Non-uniform Circuit Lower Bounds - P
art I
DTSTART;VALUE=DATE-TIME:20101130T083000Z
DTEND;VALUE=DATE-TIME:20101130T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-995@cern.ch
DESCRIPTION:Ryan Williams\, a postdoc at IBM Almaden\, posted a manuscript
about a week ago on his home page (http://www.cs.cmu.edu/~ryanw/) proving
that bounded depth circuits with AND\, OR and MOD-m gates (also called AC
C circuits) are not powerful enough to compute all of non-deterministic ex
ponential time (NEXP). This result appears to have created quite a buzz on
the theory blogs. I plan to give an informal presentation on Ryan's new r
esult trying to explain what the fuss is all about.\n\nThe first part will
be accessible to a general audience. In this part\, we will try to unders
tand the statement that Ryan proves. While doing so\, I'll give a tour of
circuit complexity over the last 3 decades\, mentioning some of its succes
ses\, its setbacks and how Ryan's result resurrects some of the hope we or
iginally had for circuit complexity.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=995
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=995
END:VEVENT
BEGIN:VEVENT
SUMMARY:Ryan Williams' New Results on Non-uniform Circuit Lower Bounds - P
art II
DTSTART;VALUE=DATE-TIME:20101130T103000Z
DTEND;VALUE=DATE-TIME:20101130T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-996@cern.ch
DESCRIPTION:Ryan Williams\, a postdoc at IBM Almaden\, posted a manuscript
about a week ago on his home page (http://www.cs.cmu.edu/~ryanw/) proving
that bounded depth circuits with AND\, OR and MOD-m gates (also called AC
C circuits) are not powerful enough to compute all of non-deterministic ex
ponential time (NEXP). This result appears to have created quite a buzz on
the theory blogs. I plan to give an informal presentation on Ryan's new r
esult trying to explain what the fuss is all about.\n\nThe second part wil
l be more technically involved. In this part\, we will brave ourselves and
dwell into Ryan's proof of NEXP not contained in ACC.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=996
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=996
END:VEVENT
BEGIN:VEVENT
SUMMARY:Modelling and Analysis of Real Time Systems with Logic Programming
and Constraints
DTSTART;VALUE=DATE-TIME:20101201T060000Z
DTEND;VALUE=DATE-TIME:20101201T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-992@cern.ch
DESCRIPTION:Embedded systems are increasingly being deployed in a wide var
iety of applications. Most\, if not all\, of these applications involve an
electronic controller with discrete behaviour controlling a continuously
evolving plant. Because of their hybrid behaviour (discrete and continuous
) and reactive behaviour\, the formal verification of embedded systems pos
e new challenges.\n\nIn my PhD dissertation\, on which this talk is based\
, proposed are two techniques to verify Linear Hybrid Automata (LHA) model
s\, namely: (a) reachability analysis – where the technique of static pr
ogram analysis and the theory of abstract interpretation are applied and (
b) abstract model checking - where the technique of model checking and the
theory of abstract interpretation are synergistically applied. We impleme
nted these techniques resulting in a systematic framework to verify LHA mo
dels with respect to properties specified in the language of Computation T
ree Logic (CTL). In this framework: (i) an LHA model is encoded in the lan
guage of Constraint Logic Programming (CLP)\, (ii) reachable state space o
f an LHA is computed as the minimum model of the equivalent CLP encoding.
So computed reachable state space forms the basis to do reachability analy
sis and abstract model checking.\n\nWe demonstrate the applicability of th
e proposed techniques/framework with examples taken from the literature.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=992
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=992
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rationality of Galois to Irrationality of Nash
DTSTART;VALUE=DATE-TIME:20101203T113000Z
DTEND;VALUE=DATE-TIME:20101203T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1010@cern.ch
DESCRIPTION:I will talk about a simple and interesting way to link Évaris
te Galois(1811-1832) and John Nash(1928- ). I will also present some issue
s that I encountered during my explorations while establishing this link.\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1010
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1010
END:VEVENT
BEGIN:VEVENT
SUMMARY:Polynomial Learning of Distribution Families
DTSTART;VALUE=DATE-TIME:20101203T060000Z
DTEND;VALUE=DATE-TIME:20101203T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-978@cern.ch
DESCRIPTION:The study of Gaussian mixture distributions goes back to the l
ate 19th century\, when Pearson introduced the method of moments to analyz
e the statistics of a crab population. They have since become one of the m
ost popular tools of modeling and data analysis\, extensively used in spee
ch recognition\, computer vision and other fields. Yet their properties ar
e still not well understood.\n\nIn my talk I will discuss some theoretical
aspects of the problem of learning Gaussian mixtures. In particular\, I w
ill discuss our recent result with Mikhail Belkin\, which\, in a certain s
ense\, completes work on an active recent topic in theoretical computer sc
ience by establishing quite general conditions for polynomial learnability
of mixture distributions.\n\nhttps://indico.tifr.res.in/indico/conference
Display.py?confId=978
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=978
END:VEVENT
BEGIN:VEVENT
SUMMARY:Breathe to Stay Cool: Adjusting Cell Sizes to Reduce Energy Consum
ption
DTSTART;VALUE=DATE-TIME:20101203T090000Z
DTEND;VALUE=DATE-TIME:20101203T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-983@cern.ch
DESCRIPTION:Reducing the energy consumption of a wireless cellular network
is an important and urgent problem. We studied the effect of cell sizes o
n the energy consumed by the network\, assuming base station technologies
of today and the future. Making cell sizes too small or too large can sign
ificantly increase energy consumption. We show that the optimal cell size
from an energy perspective depends on a number of factors\, including base
station technology\, data rates\, and traffic demands. Given that traffic
varies significantly during a day\, dynamically adjusting cell sizes can
help reduce energy consumption. We propose a practical\, 2-level scheme th
at adjusts cell sizes between two fixed values\, and show an energy saving
of up to 40%. The talk also presents some self-organizing techniques to a
llow this dynamic cell size adjustment.\n\nI will provide a brief introduc
tion to Bell Labs India\, and cover some other research projects currently
underway. I will also be happy to speak individually or in small groups a
fter the presentation with any students who’d like to learn more about B
ell Labs.\n\nBio: Girija Narlikar is a Distinguished Member of Staff at Be
ll Labs. She has spent several years at Bell Labs USA and now Bell Labs In
dia\, on developing new technologies in the areas of wireless network infr
astructure\, high-speed packet processing\, network security\, and mobile
applications. She has an MS and PhD from CMU\, and a BTech from IIT Bombay
\, all in Computer Science. She has spent most of her early years in TIFR\
, and has very fond memories of the place!\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=983
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=983
END:VEVENT
BEGIN:VEVENT
SUMMARY:Linear System Over Finite Abelian Groups
DTSTART;VALUE=DATE-TIME:20101207T103000Z
DTEND;VALUE=DATE-TIME:20101207T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-999@cern.ch
DESCRIPTION:We consider a system of linear constraints over any finite abe
lian group $G$ of the following form: $ell_i(x_1\,ldots\,x_n) equiv ell_{i
\,1}x_1+cdots+ell_{i\,n}x_n in A_i$ for $i=1\,ldots\,t$ and each $A_i subs
et G$\, $ell_{i\,j}$ is an element of $G$ and $x_i$'s are Boolean variable
s. Our main result shows that the subset of the Boolean cube that satisfie
s these constraints has exponentially small correlation with the $text{MOD
}_q$ boolean function\, when the order of $G$ and $q$ are co-prime numbers
.\n\nOur work extends the recent result of Chattopadhyay and Wigderson (FO
CS'09) who obtain such a correlation bound for linear systems over cyclic
groups whose order is a product of two distinct primes or has at most one
prime factor. Our result also immediately yields the first exponential bou
nds on the size of boolean depth-three circuits of the form $text{MAJ} cir
c text{AND} circ text{MOD}_m^A$ for computing the $text{MOD}_q$ function\,
when $m\,q$ are co-prime. No superpolynomial lower bounds were known for
such circuits for computing any explicit function\, when $m$ had three or
more distinct prime factors.\n\nThis completely solves an open problem pos
ed by Beigel and Maciel (Complexity'97) (this is joint work with Shachar L
ovett\, IAS-Princeton).\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=999
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=999
END:VEVENT
BEGIN:VEVENT
SUMMARY:Formal Proofs: From Principia Mathematica to Automated Reasoning
DTSTART;VALUE=DATE-TIME:20101208T060000Z
DTEND;VALUE=DATE-TIME:20101208T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1023@cern.ch
DESCRIPTION:The year 2010 is the centenary of Whitehead and Russell's `Pri
ncipia Mathematica'. With the advent of computer science the notion of for
mal proofs from `Principia' have evolved from a mere theoretical curiosity
to something that is possible in practice. Recent advances in the field o
f `Interactive Proof Checking' with the associated development of powerful
tools such as `Proof Assistants' have given rise to an interesting conseq
uence -- viz. the practical feasibility of importing techniques developed
in the computer science community and redeploying them to improve the main
activity of the working mathematician\, namely the process of proof devel
opment. At the core of such redeployed techniques lie the notions of forma
l systems\, formal reasoning\, and formal proofs. However the process of f
ormalizing mathematics is a highly non-trivial task\, and gives rise to a
number of challenging and interesting issues. This talk will give an overv
iew of related research in the area of automated reasoning.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=1023
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1023
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum Entanglement
DTSTART;VALUE=DATE-TIME:20101210T083000Z
DTEND;VALUE=DATE-TIME:20101210T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1027@cern.ch
DESCRIPTION:Entanglement is one of the most striking features of quantum m
echanics. The EPR paradox and Bell's inequality are some of the famous exa
mples where it manifests itself. I start by defining entanglement for a bi
partite system and talk about various measures of entanglement. Some of th
e other topics covered are: entanglement witness\, relative entropy of ent
anglement\, entanglement of formation and squashed entanglement.\n\nhttps:
//indico.tifr.res.in/indico/conferenceDisplay.py?confId=1027
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1027
END:VEVENT
BEGIN:VEVENT
SUMMARY:Boltzmann Sampling
DTSTART;VALUE=DATE-TIME:20101213T083000Z
DTEND;VALUE=DATE-TIME:20101213T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-922@cern.ch
DESCRIPTION:Uniform random generation is a central issue in combinatorics\
, with applications in many fields of computer science. Classical random s
amplers are designed to generate combinatorial structures of a given size\
; on the contrary\, under the Boltzmann model\, objects are generated with
a randomly varying size\, which allows for the design of particularly eff
icient samplers. The aim of this talk is to give an overview of Boltzmann
method\, from the original theoretical framework to effective random sampl
ers and their applications\, including recent developments.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=922
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=922
END:VEVENT
BEGIN:VEVENT
SUMMARY:Two Time Scale Algorithms for Variance-penalized Control and Risk-
neutral Semi-Markov Control
DTSTART;VALUE=DATE-TIME:20101216T090000Z
DTEND;VALUE=DATE-TIME:20101216T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1005@cern.ch
DESCRIPTION:The two time scale framework has been applied in numerous sett
ings for solving Markov decision processes. It has some remarkable propert
ies that allow it to develop solution algorithms for problems that are dif
ficult to solve with single-time-scale algorithms\, such as classical valu
e iteration or Q-Learning. In this talk\, we will discuss two applications
of this framework. The first will be for solving a variance-penalized Mar
kov decision process using dynamic programming. The second application wil
l be for developing an actor critic algorithm that can solve a risk-neutra
l semi-Markov decision process. For the actor critic\, we will present som
e numerical results from a case study in airline revenue management (this
is joint work with Sean Meyn of the University of Illinois and Susan Murra
y\, Ketaki Kulkarni\, and Katie Grantham of Missouri S & T).\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=1005
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1005
END:VEVENT
BEGIN:VEVENT
SUMMARY:Compressed Sensing: A Layman View
DTSTART;VALUE=DATE-TIME:20101216T053000Z
DTEND;VALUE=DATE-TIME:20101216T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1030@cern.ch
DESCRIPTION:Compressed sensing is a novel paradigm for signal acquisition\
, proposed recently by Candes et al\, Donoho. As the dust around the initi
al rapid developments settles\, a more simple and unified structure emerge
s\, which is surprisingly linear. This talk provides an introductory expos
ition\, with simple examples showing the cohesive picture.\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=1030
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1030
END:VEVENT
BEGIN:VEVENT
SUMMARY:Induction\, Invariants\, and Abstraction
DTSTART;VALUE=DATE-TIME:20101220T060000Z
DTEND;VALUE=DATE-TIME:20101220T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1043@cern.ch
DESCRIPTION:Invariants play a key role in verifying properties of imperati
ve programs. Inductive reasoning is essential to verifying properties of r
ecursive programs. Relationship between derivation of loop invariants and
speculation of lemmas in inductive reasoning is explored. Abstraction is a
n effective heuristic for approximating program behavior in order to deriv
e invariants. Interaction among developing abstractions\, inductive reason
ing\, and generating invariants is investigated.\n\nBio: Deepak Kapur is a
distinguished professor of computer science at the University of New Mexi
co at Albuquerque. From 1998 until 2006\, he served as the chair of the co
mputer science department there. He has conducted research in areas of aut
omated deduction\, induction theorem proving\, term rewriting\, unificatio
n theory\, formal methods\, algebraic and geometric reasoning and their ap
plications. His group built one of the first rewrite-based theorem provers
\, called Rewrite Rule Laboratory. He served as the editor-in-chief of the
Journal of Automated Reasoning from 1993-2007. He is on the editorial boa
rd of Journal of Symbolic Computation and other journals. Last year he rec
eived the Herbrand Award for distinguished contributions to automated reas
oning.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=10
43
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1043
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Computational Framework for Exploring the Role of Speech Product
ion in Speech Processing/Recognition
DTSTART;VALUE=DATE-TIME:20101222T083000Z
DTEND;VALUE=DATE-TIME:20101222T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1033@cern.ch
DESCRIPTION:It has been shown several times that the speech recognition ac
curacy improves if the direct measurement of speech articulation is used i
n addition to the speech acoustics from the talker. However\, access to su
ch direct speech articulation data during speech processing/recognition is
not feasible in practice. In my presentation\, I shall show that the spee
ch production/articulation features can be estimated from the speech signa
l of any arbitrary talker although these features are not directly availab
le from the talker. For the estimation of production-oriented features\, I
shall talk about a talker-independent acoustic-to-articulatory inversion
framework using generalized smoothness criterion\, which requires parallel
articulatory and acoustic data from a single subject only and this traini
ng subject need not be any of the talkers. Use of these estimated features
improves the acoustic-feature based recognition accuracy by ~4% (absolute
) in a phonetic recognition experiment on TIMIT corpus. Interestingly\, wh
en the training subject is interpreted as a listener\, the production-orie
nted features and\, hence\, the speech recognition can be interpreted as l
istener-specific. We will see that such a listener-specific framework to s
peech processing/recognition provides a production-oriented explanation of
the variability in recognition accuracy by non-native listeners. We will
also see that the listener-specific framework acts as a bridge between the
scientific and technological viewpoints towards the role of speech produc
tion in speech perception in the human speech communication.\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=1033
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1033
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Theory of Graph Games: Mixing Chess\, Soccer and Poker
DTSTART;VALUE=DATE-TIME:20101223T060000Z
DTEND;VALUE=DATE-TIME:20101223T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1045@cern.ch
DESCRIPTION:In this talk we will consider various classes of games played
on graphs with the classical objectives that are studied in the theory of
verification. We will present a survey of results for various different cl
asses of game graphs and show how more general class of games leads to mor
e complicated results. We will start with some basic and fundamental resul
ts in game theory on graphs\, and then show how they can be generalized to
more different class of games.\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=1045
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1045
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Voting Machine Design for Verification and Testability
DTSTART;VALUE=DATE-TIME:20101228T060000Z
DTEND;VALUE=DATE-TIME:20101228T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1058@cern.ch
DESCRIPTION:I will present an approach for the design and analysis of an e
lectronic voting machine based on a novel combination of formal verificati
on and systematic testing by humans. The system was designed specifically
to enable verification and testing. In our architecture\, the voting machi
ne is a finite-state transducer that implements the bare essentials requir
ed for an election. We formally specify how each component of the machine
is intended to work and formally verify that an implementation of our desi
gn meets this specification. However\, it is more challenging to verify th
at the composition of these components will behave as a voter would expect
\, because formalizing human expectations is difficult. We show how system
atic testing by humans can be used to address this issue\, and in particul
ar to verify that the machine will behave correctly on election day. I wil
l conclude with some observations on what we learned from this project and
some thoughts on how the ideas might generalize.\n\nBio: Available at htt
p://www.eecs.berkeley.edu/~sseshia/bio.html\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=1058
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1058
END:VEVENT
BEGIN:VEVENT
SUMMARY:Latent Dirichlet Allocation forText Segmentation
DTSTART;VALUE=DATE-TIME:20101228T103000Z
DTEND;VALUE=DATE-TIME:20101228T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-935@cern.ch
DESCRIPTION:n this presentation\, first we visit latent Dirichlet allocati
on (LDA)\, an unsupervised topic model\, and propose its application for t
he task of text segmentation. The proposed methodology has state-of-the-ar
t performance on a benchmark database\, is able to perform segmentation in
an online manner\, and assigns a meaningful topic distribution to each se
gment. The last point is particularly interesting for information retrieva
l at segment level. Another important discussion will be on how the comput
ational cost associated with the dynamic programming (DP) algorithm typica
lly used for the search can be reduced by a factor of more than 95%\, and
the usability of this result to the entire domain of text segmentation.\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=935
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=935
END:VEVENT
BEGIN:VEVENT
SUMMARY:Representation Discovery for High-Dimensional Data Mining and Cont
rol
DTSTART;VALUE=DATE-TIME:20110110T103000Z
DTEND;VALUE=DATE-TIME:20110110T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1050@cern.ch
DESCRIPTION:The explosive growth in digital archiving poses many fundament
al challenges in machine learning. This talk will describe a confluence of
recent ideas from applied mathematics\, statistics\, and machine learning
on constructing low-dimensional representations that capture the intrinsi
c statistical properties of the original data. I will present new represen
tation discovery algorithms motivated by two application domains: cross-li
ngual information retrieval and robotics. I will describe new alignment al
gorithms that find correspondences across languages by projecting document
collections onto a common low-dimensional manifold. I also describe new r
einforcement learning algorithms for solving robot control problems modele
d as Markov decision processes that dynamically construct sparse represent
ations for approximating value functions.\n\nBIO: Sridhar Mahadevan is a p
rofessor in the Computer Science Department at the University of Massachus
etts\, Amherst\, where he co-directs the Autonomous Learning Laboratory. P
rofessor Mahadevan's research interests span several areas of artificial i
ntelligence and computer science\, including machine learning\, decision m
aking\, and robotics. He has published over a hundred articles in these ar
eas\, as well as three books.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=1050
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1050
END:VEVENT
BEGIN:VEVENT
SUMMARY:Coloring\, Embedding\, Compression and Data Structure Problems on
Uniform Hypergraphs
DTSTART;VALUE=DATE-TIME:20110111T083000Z
DTEND;VALUE=DATE-TIME:20110111T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1064@cern.ch
DESCRIPTION:We study four problems on uniform hypergraph. First\, we prese
nt a streaming algorithm for two-coloring uniform hypergraphs with limited
number of hyperedges. Second\, we show relationships between four differe
nt geometric parameters of the embedding of a complete 3-uniform hypergrap
h in 3-dimensional space. Third\, we present a compression scheme for unif
orm hypertrees\, which are generalizations of trees. Finally\, we present
a data structure for succinctly storing a hyperedge of a complete uniform
hypergraph in order to support membership queries in the bitprobe model.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1064
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1064
END:VEVENT
BEGIN:VEVENT
SUMMARY:High Throughput Network Analysis
DTSTART;VALUE=DATE-TIME:20110113T103000Z
DTEND;VALUE=DATE-TIME:20110113T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1012@cern.ch
DESCRIPTION:Many real-world systems are naturally represented as graphs or
networks\, and a variety of techniques and measures exist for their analy
sis. However\, studies of networks typically employ only a small\, largely
arbitrary subset of these\, and the lack of a systematic comparison makes
it unclear which metrics are redundant or complementary. We present a fra
mework for systematic analysis of networks and network metrics\, and use i
t to analyse a large and diverse set of real networks\, along with several
kinds of synthetic model-generated networks\, making use of nearly four h
undred network metrics or summary statistics thereof. We demonstrate the u
tility of the framework for finding redundant metrics\, fitting models to
real networks\, classification of networks\, studying evolving networks\,
relating network features to evolutionary phylogenies\, and determining th
e robustness of metrics to network damage and sampling effects.\n\nhttps:/
/indico.tifr.res.in/indico/conferenceDisplay.py?confId=1012
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1012
END:VEVENT
BEGIN:VEVENT
SUMMARY:Simulation-based Computation of the Correlation Function in a Levy
-driven Queue
DTSTART;VALUE=DATE-TIME:20110118T103000Z
DTEND;VALUE=DATE-TIME:20110118T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1034@cern.ch
DESCRIPTION:In this talk I consider a single-server queue with Levy input\
, and in particular its workload process $Q(t)$\, focusing on its correlat
ion structure. With the correlation function defined as $r(t) := {rm Cov}(
Q(0)\, Q(t))/{rm Var} Q(0)$ (assuming the workload process is in stationar
ity at time 0)\, we first study its transform $int_0^infty r(t)e^{-theta t
} dt\,$ both for the case that the Levy process has positive jumps\, and t
hat it has negative jumps. These expressions allow us to prove that $r(t)$
is positive\, decreasing\, and convex\, relying on the machinery of compl
etely monotone functions. For the light-tailed case\, we estimate the beha
vior of $r(t)$ for $t$ large. We then focus on techniques to estimate $r(t
)$ by simulation. Naive simulation techniques require roughly $1/r(t)^2$ r
uns to obtain an estimate of a given precision\, but we develop a coupling
technique that leads to substantial variance reduction (required number o
f runs being roughly $1/r(t)$). If this is augmented with importance sampl
ing\, it even leads to a logarithmically efficient algorithm.\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=1034
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1034
END:VEVENT
BEGIN:VEVENT
SUMMARY:Information Flow in Wireless Networks: From Matroids to Cell Phone
s
DTSTART;VALUE=DATE-TIME:20110118T053000Z
DTEND;VALUE=DATE-TIME:20110118T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1088@cern.ch
DESCRIPTION:We consider a wireless relay network where many relay nodes he
lp communication between a source and a destination node. The goal is to u
nderstand fundamental information-theoretic limits and efficient communica
tion architectures in such networks. Wireless communication network is cha
racterized by the superposition and broadcast nature of signals\, in addit
ion to the thermal background noise. Even (apparently) simple-looking netw
orks\, are analytically intractable.\n\nRecently\, a linear deterministic
network was proposed\, which captures the signal-interaction in wireless n
etworks while downplaying the role of noise\; making the problem analytica
lly tractable\, but at the same time promising useful insights into the wi
reless network. Algebraic and combinatorial tools from Matroid theory have
been applied to characterize the flow of information in such networks\; w
here the flow captures the amount of information flowing through a relay n
ode. In this talk\, we attempt to lift the insights from these results bac
k to the wireless network. (A detailed manuscript on this work can be obta
ined from http://arxiv.org/abs/1012.0416)\n\nBiography:\n\nAdnan Raja is c
urrently pursuing his PhD in the ECE Department at the University of Illin
ois at Urbana-Champaign. He obtained his B.Tech in EE from IIT Bombay in 2
006 and MS in ECE from UIUC in 2008. He is currently a visiting student at
IISc\, Bangalore.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=1088
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1088
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum Entanglement
DTSTART;VALUE=DATE-TIME:20110125T103000Z
DTEND;VALUE=DATE-TIME:20110125T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1108@cern.ch
DESCRIPTION:Entanglement is one of the most striking features of quantum m
echanics and it gives rise to situations that are not possible classically
. For example\, a subsystem can have more disorder than the whole system.
Entanglement gives rise to some special applications such as quantum telep
ortation and superdense coding. This project is a study of entanglement. W
e define quantum entanglement and then study how to distinguish entangled
and non-entangled (separable) states. To accomplish the latter goal\, we s
tudy the concept of entanglement witness and the positive partial transpos
e (PPT) criterion. Then we study the disorder in composite systems when th
ey are en- tangled and unentangled. In the end\, we briefly look at some m
easures of entanglement such as the relative entropy of entanglement.\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1108
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1108
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Small World Phenomenon
DTSTART;VALUE=DATE-TIME:20110128T113000Z
DTEND;VALUE=DATE-TIME:20110128T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1116@cern.ch
DESCRIPTION:The experience of meeting a complete stranger with whom we hav
e apparently little in common and ﬁnding unexpectedly that we share a mu
tual acquaintance is one with which most of us are familiar—“It’s a
small world!” we often say. In the 1960s\, noted psychologist Stanley Mi
lgram quantified this small world phenomenon by demonstrating that randoml
y selected individuals from almost anywhere in the United States were “c
onnected” via a chain of no more than six intermediate acquaintances. In
this talk\, we will analyse the social network models proposed by “Watt
s & Strogatz” and “Klienberg” that attempt to explain this phenomeno
n.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1116
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1116
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Unified Framework for Testing Linear-Invariant Properties
DTSTART;VALUE=DATE-TIME:20110131T063000Z
DTEND;VALUE=DATE-TIME:20110131T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1084@cern.ch
DESCRIPTION:In a sequence of recent papers\, Sudan and coauthors have inve
stigated the relation between testability of properties of Boolean functio
ns and the invariance of the properties with respect to transformations of
the domain. Linear-invariance is arguably the most common such symmetry f
or natural properties of Boolean functions on the hypercube. Hence\, it is
an important goal to find necessary and sufficient conditions for testabi
lity of linear-invariant properties. This is explicitly posed as an open p
roblem in a recent survey of Sudan. We obtain the following results:\n\n1.
We show that every linear-invariant property that can be characterized by
forbidding induced solutions to a (possibly infinite) set of linear equat
ions can be tested with one-sided error.\n\n2. We show that every linear-i
nvariant property that can be tested with one-sided error can be character
ized by forbidding induced solutions to a (possibly infinite) set of syste
ms of linear equations.\n\nWe conjecture that our result from item (1) can
be extended to cover systems of linear equations. We further show that th
e validity of this conjecture would have the following implications:\n\n1.
It would imply that every linear-invariant property that is closed under
restrictions to linear subspaces is testable with one-sided error. Such a
result would unify several previous results on testing Boolean functions\,
such as the results on testing low-degree\npolynomials and results on tes
ting Fourier dimensionality.\n\n2. It would imply that a linear-invariant
property P is testable with one-sided error *if and only if* P is closed u
nder restrictions to linear subspaces\, thus resolving Sudan's problem (jo
int work with Elena Grigorescu and Asaf Shapira).\n\nhttps://indico.tifr.r
es.in/indico/conferenceDisplay.py?confId=1084
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1084
END:VEVENT
BEGIN:VEVENT
SUMMARY:Specification and Verification of Timed and Communicating Systems
DTSTART;VALUE=DATE-TIME:20110131T103000Z
DTEND;VALUE=DATE-TIME:20110131T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1115@cern.ch
DESCRIPTION:Our goal is to use formal methods to reason about systems wher
e time and concurrency play a significant role. We are interested in check
ing if the behaviours exhibited by an implementation conform to those stip
ulated by the specification in a timed and distributed system.\n\nTo descr
ibe the behaviours of distributed systems which operate on a global time\,
we introduce concrete and abstract notions of message sequence charts (MS
Cs) with timing. For appropriate formalisms of implementation (timed messa
ge passing automata) and specification (monadic second order logic) over t
imed MSCs\, we obtain an expressive equivalence result. Infinite collectio
ns of MSCs with timing can also be specified as message sequence graphs wi
th timing. We address two natural problems that arise in this setting.\n\n
Finally\, we consider an alternate system model where clocks in different
components of a distributed system evolve at different local rates. We exa
mine different semantics that allow us to check for good and bad behaviour
s in the system and show undecidability as well as regularity results.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1115
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1115
END:VEVENT
BEGIN:VEVENT
SUMMARY:Formal Definitions and Complexity Results for Trust Relations and
Trust Domains
DTSTART;VALUE=DATE-TIME:20110201T103000Z
DTEND;VALUE=DATE-TIME:20110201T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1094@cern.ch
DESCRIPTION:We propose computational\, declarative definitions of the conc
epts of weak and strong trust relations between interacting agents\, and t
rust domains of trust-related agents in distributed systems. Our definitio
ns yield computational complexity results for deciding potential and actua
l trust relationships and membership in trust domains\, as well as a posit
ive (negative) compositionality result for strong (weak) trust domains. Ou
r defining principle for weak and strong trust is (common) belief in and k
nowledge of agent correctness\, respectively.\n\nKEYWORDS:\n\ncomputabilit
y\, compositionality\, and scalability of trust and trustworthiness\; comp
uter-aided decision making (CADM)\; dependable distributed or multi-agent
systems\; modal logics of belief and knowledge.\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=1094
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1094
END:VEVENT
BEGIN:VEVENT
SUMMARY:Nearly Tight Bounds for Testing Function Isomorphism
DTSTART;VALUE=DATE-TIME:20110202T090000Z
DTEND;VALUE=DATE-TIME:20110202T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1061@cern.ch
DESCRIPTION:We study the problem of testing structural equivalence (isomor
phism) between a pair of Boolean functions $f\,g:{0\,1}^n to {0\,1}$. Our
main focus is on the most studied case\, where one of the functions is giv
en (explicitly)\, and the other function can be queried.\n\nWe prove that
for every $k leq n$\, the query complexity of testing isomorphism to $k$-j
untas is $Omega(k)$ and $O(k log k)$. In particular\, the (worst-case) que
ry complexity of testing isomorphism to a given function $f:{0\,1}^n to {0
\,1}$ is $widetildeTheta (n)$.\n\nThus our bounds are nearly tight. Our lo
wer bound and upper bound results improves the known bound obtained by Fis
cher et al. (2004)\, Blais and O'Donnell (2010)\, and recently by Alon and
Blais (2010).\n\nOur proof can also be extended to give polynomial query-
complexity lower bounds for the problems of testing whether a function has
a circuit of size $\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=1061
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1061
END:VEVENT
BEGIN:VEVENT
SUMMARY:Paradoxes\, Computers\, and Reproduction
DTSTART;VALUE=DATE-TIME:20110203T083000Z
DTEND;VALUE=DATE-TIME:20110203T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1126@cern.ch
DESCRIPTION:How can one write a computer program that outputs itself? Is t
he sentence "this sentence is false" true or false? Can one write a progra
m that will distinguish between programs that have infinite loops\, and th
ose that do not? Is it possible to make a machine that creates a copy of i
tself\, or even something more complex than itself? How does one show that
there are more real numbers than natural numbers? In 1969\, Lawvere showe
d that the answers to all these questions (Kleene's recursion theorem\, li
ar's paradox\, Turing's halting problem\, von Neumann's self-reproducing a
utomata\, Cantor's diagonalization argument\, and\, of course\, Godel's fi
rst incompleteness theorem) follow from one very short result in the setti
ng of "cartesian closed categories." I will present this result of Lawvere
. If interested\, you may join us for the discussion that follows.\n\nPrer
equisites: An intuitive notion of what is a computer and what is a compute
r program should suffice.\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=1126
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1126
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invariance Principles in Probability and Their Applications in The
oretical Computer Science
DTSTART;VALUE=DATE-TIME:20110208T103000Z
DTEND;VALUE=DATE-TIME:20110208T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1143@cern.ch
DESCRIPTION:In recent years\, invariance principles in probability (centra
l limit theorem\, Berry-Esseen theorem\, and their high degree and multi-d
imensional generalizations) have come very useful in\ntheoretical computer
science\, particularly in the areas of hardness of approximation\, derand
omization\, learning theory\, social choice and property testing.\n\nI'll
begin this talk by insulting your intelligence by recalling a simple proof
of the central limit theorem\, with rather weak error bounds using the hy
brid argument (aka the Lindeberg method). We will then see how this simple
proof can be generalized to yield the high-degree and multi-dimensional v
ersions. Time permitting\, I will discuss some of the applications of thes
e generalizations to theoretical computer science.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=1143
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1143
END:VEVENT
BEGIN:VEVENT
SUMMARY:Chemical Reaction Networks
DTSTART;VALUE=DATE-TIME:20110210T083000Z
DTEND;VALUE=DATE-TIME:20110210T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1155@cern.ch
DESCRIPTION:An elaborate symphony is orchestrated in every living cell. Th
e score is written in DNA\, and is played out by RNA and protein enzymes\,
but where is the conductor of this symphony? Research in molecular biolog
y over the past half century suggests that it is the intricate biochemical
circuits (gene regulatory networks\, cell signalling pathways\, etc.) tha
t play this role. In this context\, it may be of value to learn to read an
d design the logic of biochemical circuits. I will give a tutorial-style i
ntroduction to the mathematics of chemical reaction networks\, with a spec
ial emphasis on mass action kinetics.\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=1155
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1155
END:VEVENT
BEGIN:VEVENT
SUMMARY:Resource Constrained Signal Processing Algorithms and Architecture
s: A Holistic Approach
DTSTART;VALUE=DATE-TIME:20110211T083000Z
DTEND;VALUE=DATE-TIME:20110211T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1129@cern.ch
DESCRIPTION:Current emerging applications in the areas of Wireless sensor
networks\, person-centric remote health monitoring and pervasive computing
require sophisticated signal processing algorithms and their correspondin
g architectural implementation with low complexity\, limited area and low
power consumption. Being computationally intensive\, direct architectural
mapping of the traditional signal processing algorithms may not be suitabl
e for such resource constrained applications. It necessitates further inve
stigation into the contemporary signal processing techniques from the pers
pective of algorithm-architecture holistic optimization approach so that i
t would become suitable for such environment. This is the prime focus of t
his presentation. However an attempt will also be made to add a bigger per
spective to the afore mentioned signal processing challenge so that it wil
l go beyond the boundary of electrical and electronic engineering and open
up opportunities of multidisciplinary research where engineering science\
, physics\, chemistry\, mathematics and biology will all be amalgamated to
gether. To make this presentation more interactive and interesting\, main
focus will be given on explaining the underlying concepts of the proposed
techniques with the help of diagrams and with the minimal use of mathemati
cs. Finally the presentation will end highlighting the author’s ongoing
research and vision for near future.\n\nAuthor’s Bio-data: Amit Acharyya
received the B.E. degree in Electronics and Communication Engineering fro
m National Institute of Technology\, Durgapur\, India\, in 2005 and joined
National Centre for Radio Astrophysics\, Tata Institute of Fundamental Re
search\, Pune\, India as a visiting researcher. In September\, 2005 he joi
ned Defence Research and Development Organisation (DRDO) as a scientist an
d worked in the VLSI Division of Advanced Numerical Research and Analysis
Group (ANURAG) Laboratory at Hyderabad\, India. In July\, 2007 he joined P
ervasive Systems Centre research group in the School of Electronics and Co
mputer Science in the University of Southampton\, U.K.\, as a visiting res
earcher and soon after he joined the same research group to pursue PhD. He
submitted his PhD Thesis successfully in December\, 2010 and presently he
is a Research Fellow in the University of Southampton\, UK. His research
interests include signal processing algorithms\, VLSI architectures\, low
power design techniques\, computer arithmetic\, numerical analysis\, linea
r algebra\, genomic and proteomic signal processing\, bio-informatics and
electronic aspects of pervasive computing.\n\nWebpage: http://www.ecs.soto
n.ac.uk/people/aa07r\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=1129
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1129
END:VEVENT
BEGIN:VEVENT
SUMMARY:Non-euclidean Algebra
DTSTART;VALUE=DATE-TIME:20110214T103000Z
DTEND;VALUE=DATE-TIME:20110214T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1131@cern.ch
DESCRIPTION:Euclidean geometry today is most naturally understood as affin
e geometry equipped with a suitable bilinearity as a basis for notions of
length\, angle\, area\, rectangles\, circles\, etc. In this talk we expres
s the affine content of Euclid's Postulates 2\, 5\, and 1 algebraically in
that order. In our account Euclid's Fifth or Parallel Postulate is expres
sed as an equation parametrized with the number 6. Parameters in the range
3 to 5 yield discrete notions of elliptical geometry while 7 and beyond d
o the same for hyperbolic geometry. Had the tools of the past half century
of universal algebra been available to Euclid\, the first of his thirteen
books could have been both simpler and more general\, and the conceptual
obstacles to hyperbolic geometry overcome more easily.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=1131
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1131
END:VEVENT
BEGIN:VEVENT
SUMMARY:Topoalgebra
DTSTART;VALUE=DATE-TIME:20110215T103000Z
DTEND;VALUE=DATE-TIME:20110215T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1121@cern.ch
DESCRIPTION:20th century logic has founded mathematics on the binary relat
ion of set membership\, formalized with the axioms of Zermelo and Fraenkel
. At mid-century algebra reacted by proposing instead to found mathematics
on the associative binary operation of function composition\, formalized
as category theory. But category theory then fell into the trap of making
categories themselves algebraic by introducing functors and natural transf
ormations\, which led to 2-categories and thence to n-categories and omega
-categories. In this talk we avoid this trap while simultaneously accounti
ng for the compatibility of topology and algebra in terms of the arbitrari
ly selected free and co-free objects of any category whatsoever. Applicati
ons include the topoalgebraic specification of the notion of an acyclic gr
aph\, an extensional conception of the concept of property or attribute\,
and a demonstration that C.I. Lewis's 1927 quasi-psychological notion of q
ualia is at least mathematically consistent.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=1121
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1121
END:VEVENT
BEGIN:VEVENT
SUMMARY:DIY Climate Change
DTSTART;VALUE=DATE-TIME:20110215T060000Z
DTEND;VALUE=DATE-TIME:20110215T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1130@cern.ch
DESCRIPTION:Climate skeptics today treat climate science and policies deri
ved therefrom as a global conspiracy with the goal of fleecing the non-con
spirators\, namely the skeptics. The basic objection to the conspiracy is
that it is founded on false science. Since this fleecing seems an increasi
ngly likely outcome\, it is worth examining its logic. First\, if the cons
pirators end up greatly outnumbering the skeptics then the net proceeds of
the conspiracy will be minimal. Second\, if the public can easily see for
itself that the science is actually not false\, then the skeptics are all
owing themselves to be fleeced for no good reason. This talk focuses on DI
Y climate change\, defined as what anyone with a high school background in
physics and a modicum of common sense can easily see for themselves is ha
ppening to the climate today without having to rely on the pronouncements
and impenetrably complex models of the climate science community.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1130
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1130
END:VEVENT
BEGIN:VEVENT
SUMMARY:Catalysis in Reaction Networks
DTSTART;VALUE=DATE-TIME:20110224T083000Z
DTEND;VALUE=DATE-TIME:20110224T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1188@cern.ch
DESCRIPTION:If networks of chemical reactions are the circuits of biology
then catalysts are the switches. But which species should be called cataly
sts? Chemistry textbooks answer this question when there is a single react
ion. For an entire network of reactions \, the concept becomes more nuance
d\, and has been worked out in a recent paper (arXiv:1006.3627). I will di
scuss this notion of catalysis for reaction networks\, and illustrate it w
ith an example from the "seesaw gate" of Qian and Winfree ( http://www.dna
.caltech.edu/DNAresearch_publications.html#SeesawTheory )\, which is a mot
if for engineering large reaction networks out of DNA molecules.\n\nPrereq
uisites: You should know what a graph is\, what a monomial is\, what the g
reatest common divisor of two monomials is\, and what DNA is.\n\nurl http:
//tifrbiobytes.blogspot.com\n\nhttps://indico.tifr.res.in/indico/conferenc
eDisplay.py?confId=1188
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1188
END:VEVENT
BEGIN:VEVENT
SUMMARY:Almost Settling the Hardness of Noncommutative Determinant
DTSTART;VALUE=DATE-TIME:20110301T103000Z
DTEND;VALUE=DATE-TIME:20110301T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1208@cern.ch
DESCRIPTION:The determinant and the permanent of a matrix\, though deceivi
ngly similar in their definitions\, behave very differently with respect t
o how efficiently one can compute these quantities. The determinant of a m
atrix over a field can be easily computed via Gaussian elimination while c
omputing the permanent\, as shown by Valiant\, is at least as hard as coun
ting the number of satisfiable assignments to a Boolean formula. Given thi
s\, it is natural to ask "what makes the determinant easy and permanent ha
rd to compute?\, is commutativity\, the key to efficient determinant compu
tation?"\n\nWe show that computing the determinant of an nxn matrix whose
entries are themselves 2x2 matrices over a field is as hard as computing t
he permanent over the field (Extending the recent result of Arvind and Sri
nivasan [STOC 2010]). On the other hand\, surprisingly if one restricts th
e elements to be dxd upper triangular matrices\, then determinant can be c
omputed in poly(n^d) time. Combining this with the decomposition theorem o
f finite algebras\, we get the following dichotomy result: if A is a finit
e dimensional algebra over a finite field\, then the commutativity of A/R(
A) determines efficient determinant computation (where R(A) is the radical
of A).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1
208
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1208
END:VEVENT
BEGIN:VEVENT
SUMMARY:Classical Mechanics
DTSTART;VALUE=DATE-TIME:20110303T083000Z
DTEND;VALUE=DATE-TIME:20110303T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1216@cern.ch
DESCRIPTION:This will be a one hour crash course on some of the basic idea
s of classical mechanics: configuration space\, Lagrangians\, the principl
e of stationary action\, and Noether's theorem. A reference is the online
book http://math.ucr.edu/home/baez/classical/texfiles/2005/book/classical.
pdf . In the discussion after the talk\, we will work out some JEE physics
-style mechanics problems.\n\nThe importance of classical mechanics is har
d to exaggerate. It forms the basis of all of modern physics\, and has ins
pired many deep ideas in mathematics\, from manifolds and topology\, to Li
e groups and representation theory\, to symplectic and Kahler geometry. Cl
assical mechanics is also of obvious importance for understanding most phy
sical phenomena\, including most biological phenomena. This talk will not
develop any of these connections.\n\nPrerequisites: High school-level math
ematics and physics\, including integration by parts\, should suffice.\n\n
url http://tifrbiobytes.blogspot.com\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=1216
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1216
END:VEVENT
BEGIN:VEVENT
SUMMARY:Pseudo-Visibility Graphs
DTSTART;VALUE=DATE-TIME:20110308T113000Z
DTEND;VALUE=DATE-TIME:20110308T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1224@cern.ch
DESCRIPTION:The notion of visibility graphs can be extended to that of pse
udo-visibility graphs\, where the edges between vertices do not necessaril
y remain straight line-segments. We discuss the construction of pseudo-vis
ibility graphs which cannot be drawn on the plane as any valid visibility
graph keeping the polygonal boundary unaltered\, even though they satisfy
all the necessary combinatorial conditions known for a graph to be a visib
ility graph. This proves that the necessary conditions are not sufficient
ones.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=122
4
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1224
END:VEVENT
BEGIN:VEVENT
SUMMARY:Active Sequential Hypothesis Testing
DTSTART;VALUE=DATE-TIME:20110309T053000Z
DTEND;VALUE=DATE-TIME:20110309T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1215@cern.ch
DESCRIPTION:Active sequential hypothesis testing problem arises in a broad
spectrum of applications in cognition\, communications\, design of experi
ments\, and sensor management. In all of these applications\, a decision m
aker is responsible to take actions dynamically so as to enhance informati
on about an underlying phenomena of interest in a speedy manner while acco
unting for the cost of communication\, sensing\, or data collection. In pa
rticular\, due to the sequential nature of the problem\, the decision make
r relies on his current information state to constantly (re-)evaluate the
trade-off between the precision and the cost of various actions.\n\nIn thi
s work\, we first discuss active sequential hypothesis testing as a partia
lly observable Markov decision problem. In particular\, we provide a brief
survey of the design of experiment literature and the dynamic programming
interpretation of information utility introduced by De Groot. Using Black
well ordering\, we\, then\, connect this stochastic control theoretic noti
on of information utility to the concept of stochastic degradation and unc
ertainty reduction in information theory.\n\nFinally\, we discuss the dyna
mics and expected drift of log-likelihood\, entropy\, and probability of e
rror as well as their connection to Kullback-Leibler divergence and mutual
information in order to approximate the optimal value function (i.e. the
solutions to the DP). We then utilize these value function approximations
(lower bounds) to provide simple sequential test strategies (heuristics) w
hose performance is numerically compared to the optimal policies. Finally\
, we prove the asymptotic optimality of one class of these heuristic test
strategies and\, as a special case\, recover Burnashev's coding scheme in
the context of variable-length block coding over memoryless channels with
feedback. Time permitting\, we will compare and contrast our approach with
recent results in Bayesian active learning literature (this is joint work
with Mohammad Naghshvar and Ofer Shayevitz).\n\nBio: Tara Javidi studied
electrical engineering at Sharif University of Technology\, Tehran\, Iran
from 1992 to 1996. She received the MS degrees in electrical engineering (
systems)\, and in applied mathematics (stochastics) from the University of
Michigan\, Ann Arbor\, in 1998 and 1999\, respectively. She received her
Ph.D. in electrical engineering and computer science from the University o
f Michigan\, Ann Arbor\, in 2002.\n\nFrom 2002 to 2004\, she was an assist
ant professor at the Electrical Engineering Department\, University of Was
hington\, Seattle. She joined University of California\, San Diego\, in 20
05\, where she is currently an associate professor of electrical and compu
ter engineering.\n\nTara Javidi was a Barbour Scholar during 1999-2000 aca
demic year and received an NSF CAREER Award in 2004. Her research interest
s are in communication networks\, stochastic resource allocation\, stochas
tic control theory\, and wireless communications.\n\nhttps://indico.tifr.r
es.in/indico/conferenceDisplay.py?confId=1215
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1215
END:VEVENT
BEGIN:VEVENT
SUMMARY:Ratchets\, Pawls\, and Maxwell's Demon
DTSTART;VALUE=DATE-TIME:20110310T083000Z
DTEND;VALUE=DATE-TIME:20110310T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1223@cern.ch
DESCRIPTION:I will describe Maxwell's demon\, the Smoluchowski ratchet and
pawl\, the Szilard engine that relates information and heat\, and some wo
rk of Brillouin\, Landauer\, Bennett and others on these topics. These top
ics lie at the intersection of thermodynamics and information and computer
science\, and I will aim to bring out this connection. On the biology sid
e\, it has been suggested that the molecular motors dynein and kinesin can
be modeled as "flashing ratchets."\n\nPrerequisites: Not strictly necessa
ry\, but it will help to know a little about random variables\, Shannon en
tropy\, and the ideal gas law.\n\nUrl http://tifrbiobytes.blogspot.com/\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1223
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1223
END:VEVENT
BEGIN:VEVENT
SUMMARY:Building Certificates of Regular Expressions Equivalence
DTSTART;VALUE=DATE-TIME:20110316T083000Z
DTEND;VALUE=DATE-TIME:20110316T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1229@cern.ch
DESCRIPTION:Regular expressions are mostly known as pattern matching expre
ssions in scripting languages (Perl\, sed\, awk\, etc.). They are also the
oretically studied for their strong relationship with Automata Theory. A r
egular expression denote a language (set of words) and two regular express
ions are equivalent when they denote the same language. The equivalence pr
oblem is decidable and furthermore\, any equivalence can be proved axiomat
ically. Various such axiomatisations have flourished during the last fifty
years\, we will survey the most important ones.\n\nEvery axiomatisation c
ome together with a proof of completeness of the following form: if two re
gular expressions are equivalent then there exists proof within the axioma
tic system. In principle\, when this completeness proof is constructive on
e can extract from it an algorithm that produces a certificate of the equi
valence. We have filled the gap developping a program (written in OCaml) t
hat computes certificates. For this purpose we have designed a domain spec
ific proof system with a language of commands for composing a certificate
checkable within this system (this is a joint work with Bodhayan Roy).\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1229
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1229
END:VEVENT
BEGIN:VEVENT
SUMMARY:Compressive Sensing
DTSTART;VALUE=DATE-TIME:20110316T123000Z
DTEND;VALUE=DATE-TIME:20110316T133000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1232@cern.ch
DESCRIPTION:The compressive sensing philosophy exploits the fact that many
signals of interest have a sparse representation in some basis to provide
an efficient approach to data acquisition and recovery. In this lecture\,
we will study the fundamental theorems of this area and discuss its appli
cations in the single pixel camera\, face detection and sudoku solvers.\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1232
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1232
END:VEVENT
BEGIN:VEVENT
SUMMARY:Emergent Cooperation Amongst Selfish Agents in Minority Games
DTSTART;VALUE=DATE-TIME:20110322T090000Z
DTEND;VALUE=DATE-TIME:20110322T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1249@cern.ch
DESCRIPTION:I will discuss a variation of the minority game. There are $N$
agents. Each has to choose between one of two alternatives everyday\, and
there is reward to each member of the smaller group. The agents cannot co
mmunicate with each other\, but try to guess the choice others will make\,
based only the past history of number of people choosing the two alternat
ives. I will discuss a simple probabilistic strategy using which the agent
s acting selfishly\, and independently\, can still maximize the average nu
mber of people benefitting every day. The strategy leads to a very efficie
nt utilization of resources.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=1249
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1249
END:VEVENT
BEGIN:VEVENT
SUMMARY:Slicing a Cube in N Dimensions
DTSTART;VALUE=DATE-TIME:20110329T113000Z
DTEND;VALUE=DATE-TIME:20110329T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1281@cern.ch
DESCRIPTION:We will consider the problem of cutting the maximum number of
edges of a cube (commonly known as hypercube in n-dimensions) using a plan
e (hyperplane in n-dimensions). We will see an explicit plane which gives
the maximum cut. Also\, we will use a combinatorial lemma (a generalizatio
n of Sperner's Lemma) to prove this theorem about geometry.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=1281
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1281
END:VEVENT
BEGIN:VEVENT
SUMMARY:Globus Online: Hosted Service for High-speed\, Reliable and Secure
Data Movement
DTSTART;VALUE=DATE-TIME:20110331T060000Z
DTEND;VALUE=DATE-TIME:20110331T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1217@cern.ch
DESCRIPTION:Grid computing enables scientists to combine and share the com
puter resources from multiple administrative domains to solve complex scie
ntific problems. As a result\, Large-scale collaborative science is becomi
ng increasingly common\, involving thousands of collaborators distributed
across the globe and generating massive amounts of data that must be trans
ferred across wide area networks. In practice orchestration of such transf
ers is technically challenging. One key issue is the frequent need to expl
oit parallelism in multiple dimensions. Another is dealing with failures o
f various sorts. Globus Online is a new direction that attempts to combine
the powerful ideas and technologies of hosted services with existing data
transfer engines (in particular\, GridFTP) to provide a robust\, reliable
\, secure\, and highly monitored environment for file transfers that has p
owerful yet easy-to-use interfaces. In this talk\, I will describe the mot
ivation for developing Globus Online and demonstrate its capabilities.\n\n
Bio: Raj Kettimuthu is a Computer Scientist at Argonne National Laboratory
\, USA and a research fellow at University of Chicago. He is the technolog
y coordinator for Globus GridFTP\, most widely used production quality dat
a mover. He has published more than 40 articles in the areas of Grid/Distr
ibuted computing and high performance computing. He has served on the prog
ram committee for more than 50 international conferences and workshops. He
is a senior member of both IEEE and ACM.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=1217
LOCATION:Colaba Campus AG-66
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1217
END:VEVENT
BEGIN:VEVENT
SUMMARY:Auctions and Approximations
DTSTART;VALUE=DATE-TIME:20110405T103000Z
DTEND;VALUE=DATE-TIME:20110405T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1210@cern.ch
DESCRIPTION:Suppose we want to design an auction for selling items to a se
t of customers. Such an auction will take the valuations of the customers
as input\, and specify the allocation of the items and the payments. A cru
cial aspect of this setting is the strategic behaviors of the customers: E
ach customer is a self-interested agent\, and she is willing to manipulate
the scheme by misreporting her valuation if that results in an outcome th
at is more favorable to her. The auction has to be robust to such manipula
tions. We say an auction is truthful if no customer has an incentive to mi
sreport. The objective is to come up with a truthful auction that either g
enerates large revenue\, or maximizes the total utility of all the partici
pants.\n\nAuction Theory has a rich literature in Economics. However\, it
is difficult to characterize the optimal auction in many settings. This co
nsideration\, coupled with the advent of internet advertising where the se
arch engines generate huge amount of revenue by auctioning online ad-slots
\, has motivated the Computer Science community to revisit Auction Theory
from an algorithmic perspective. Here\, we want to design approximately op
timal auctions whose outcomes can be computed in polynomial time. I will p
resent some recent results that fall under the purview of this general res
earch agenda.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=1210
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1210
END:VEVENT
BEGIN:VEVENT
SUMMARY:Concentration of Measure in High Dimensions
DTSTART;VALUE=DATE-TIME:20110405T113000Z
DTEND;VALUE=DATE-TIME:20110405T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1290@cern.ch
DESCRIPTION:We know that the Gaussian distribution concentrates sharply ar
ound its mean\, ie. the probability mass outside a few standard deviations
decreases exponentially in the number of steps taken. Such a concentratio
n result can actually be derived for areas and volumes in higher dimension
s\, as purely geometrical facts. For instance\, for the unit sphere in n-d
imensions\, most of the volume is concentrated around every slice through
the equator. Such results have various interesting implications. For examp
le\, any Lipschitz function defined on the n-dimensional sphere is more or
less constant! I will talk about some basic concentration of measure resu
lts\, and related questions like isoperimetric inequalities\, the Brun-Min
kowski inequality and possibly something on metric embeddings\, depending
on what time permits.\n\nhttps://indico.tifr.res.in/indico/conferenceDispl
ay.py?confId=1290
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1290
END:VEVENT
BEGIN:VEVENT
SUMMARY:Thermodynamics of Computation
DTSTART;VALUE=DATE-TIME:20110407T083000Z
DTEND;VALUE=DATE-TIME:20110407T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1289@cern.ch
DESCRIPTION:\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=1289
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1289
END:VEVENT
BEGIN:VEVENT
SUMMARY:Concentration of Measure in High Dimensions (Part-II)
DTSTART;VALUE=DATE-TIME:20110412T113000Z
DTEND;VALUE=DATE-TIME:20110412T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1308@cern.ch
DESCRIPTION:Last time around we saw some interesting and non-intuitive beh
aviour of volumes and surfaces in high dimensions. Keeping in line with th
is\, we will discuss the Brun-Minkowski theorem\, and use it to prove simi
lar concentration results in various settings: surface areas on the n-dime
nsional ball\, the Gaussian Measure in R^n\, and the Hamming Cube C_n. If
time permits\, I will mention a result on metric embeddings that relies on
such a concentration of measure phenomenon.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=1308
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1308
END:VEVENT
BEGIN:VEVENT
SUMMARY:Predictive Toxicity In Silico
DTSTART;VALUE=DATE-TIME:20110413T090000Z
DTEND;VALUE=DATE-TIME:20110413T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1304@cern.ch
DESCRIPTION:Various in silico methods are employed to predict toxicity in
pharmaceutical R&D. The methods can range from simple structural alerts al
l the way to detailed mechanistic modeling of biological systems. In my ta
lk I will briefly go over some of these methods with a focus on structure-
activity relationships\, network-chemical similarity approaches and dynami
c systems modeling. The talk will cover issues around how these models are
built\, their applicability and their impact on the pharmaceutical pipeli
ne.I will discuss how the quality of predictions made influences decision-
making.\n\nAbout the Speaker: Dr. Kalyanasundaram Subramanian from Strand
life sciences\, popularly known as 'Kas'\, leads Strand's scientific and t
echnical programs and coordinates the cross-divisional efforts in R&D. Kas
's interests lie in the field of ADMET modeling and molecule design using
machine learning and systems biology techniques. He has over a decade of e
xperience in modeling biological systems.\n\nAn IIT - Bombay alumnus\, Kas
went on to complete his M.S. in Chemical Engineering from the State Unive
rsity of New York College at Buffalo. Equipped with a Ph.D. in Biomedical
Engineering from Johns Hopkins University\, Kas took up the position of Se
nior Scientist at Genetic Therapy Inc (Novartis) between 1997-2000\, where
he helped set-up a group to perform research in synthetic and hybrid vect
ors for gene delivery. Prior to Strand\, Kas headed the Collaborative R&D
group for immunology products at Entelos.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=1304
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1304
END:VEVENT
BEGIN:VEVENT
SUMMARY:Gaussian Process Modeling of Large Scale Terrain
DTSTART;VALUE=DATE-TIME:20110419T103000Z
DTEND;VALUE=DATE-TIME:20110419T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1227@cern.ch
DESCRIPTION:This talk will focus on the problem of large scale multi-senso
r multi-data-set sensor based perception and sensor fusion. The applicatio
n context of this study is large scale terrain modeling for mining automat
ion.\n\nBuilding a model of large scale terrain (5 sq km) that can adequat
ely handle uncertainty and incompleteness of sensor data in a statisticall
y sound way is a challenging problem. Most contemporary representations ar
e not equipped to model spatially correlated data and typically treat data
as being statistically independent. To obtain a comprehensive model of su
ch terrain\, typically\, multiple sensory modalities as well as multiple d
ata sets are required. This necessitates sensor fusion.\n\nIn order to add
ress these issues\, this work proposed the use of Gaussian processes (GP's
) as models of large scale terrain. The model naturally provided a multi-r
esolution representation of space\, incorporated and handled uncertainties
aptly and coped with incompleteness of sensory information. Gaussian proc
ess regression techniques were applied to estimate and interpolate (to fil
l gaps in occluded areas) elevation information across the field. A single
non-stationary (neural network) Gaussian process was shown to be powerful
enough to model large and complex terrain\, effectively handling issues r
elating to discontinuous data.\n\nExperiments were performed on large scal
e 3D data sets taken using GPS and laser scanners from a mining scenario.
Extensive statistical performance evaluation of the technique was performe
d through cross validation experiments on the aforementioned data sets. Th
ese experiments also compared the proposed modeling approach with most oth
er well known interpolation and representation methods. The outcome of the
se comparison and benchmarking experiments was that the proposed approach
will perform as well as grid based techniques or triangulated irregular ne
tworks (TIN's) for dense\, relatively flat laser scanner data sets\, howev
er\, for complex and/or sparse data sets\, the proposed Gaussian process m
odeling approach will significantly outperform grid based approaches using
most standard interpolation techniques as well as TIN's using triangle ba
sed interpolation techniques.\n\nThis work then proposed two approaches to
data fusion using Gaussian processes - one based on Heteroscedastic GP's
and the other based on Dependent GP's. The approach based on heteroscedast
ic GP's modeled the different data sets as different noisy samples of a co
mmon underlying terrain. Dependent GP based data fusion modeled each data
set using a separate GP and learnt spatial correlation information between
different GP's through auto and cross correlations. A key novelty of this
work was the derivation and use of non-stationary kernels for multi-task
problems with dependent Gaussian processes. The work based on dependent GP
's has also successfully demonstrated the simultaneous modeling/prediction
of multiple properties of the terrain (terrain elevation and color).\n\nT
o enable the approach to cope with multiple large-scale data sets\, GP app
roximations were developed for both the learning and inference stages. A l
ocal approximation method based on a "moving window" methodology and imple
mented using KD-Trees was proposed for both GP learning and inference. Fur
ther\, a block learning approach to GP learning was proposed which guarant
eed the successful use of this approach in resource constrained systems. T
hese approximation methods enabled the approach to handle large data sets\
, thereby addressing its scalability issues.\n\nABOUT THE SPEAKER:\n\nDr S
hrihari Vasudevan has a BE in Computer Science and Engineering from the Un
iversity of Madras (2002)\, an MS in Computer Science / Intelligent Roboti
cs from the University of Southern California\, USA (2004) and a DSc in In
telligent Robotics from the Swiss Federal Institute of Technology Zurich (
2008). He is currently a research fellow / lecturer at the Australian Cent
re for Field Robotics\, The University of Sydney. His research interests m
ay be summarized as the modeling and mining of sensor data. Specifically\,
he is interested in sensor based perception\, sensor fusion\, machine lea
rning and pattern recognition towards developing intelligent robots and sy
stems.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=12
27
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1227
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sum of Square Roots of Polynomials and of Integers
DTSTART;VALUE=DATE-TIME:20110419T090000Z
DTEND;VALUE=DATE-TIME:20110419T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1324@cern.ch
DESCRIPTION:Let a_1\, a_2\, ...\, a_n and b_1\, b_2\, ...\, b_n be positiv
e integers each of which is at most n bits long. Let S be the difference b
etween the sum of square roots of a_i's and the sum of square roots of the
b_j's. If S is nonzero how small can it be in absolute value? It is conje
ctured that in absolute absolue value S must be at least 1/(2^{poly(n)}).
In this talk we will see the solution of a polynomial analog of this conje
cture.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=13
24
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1324
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dyadic Rounding
DTSTART;VALUE=DATE-TIME:20110421T043000Z
DTEND;VALUE=DATE-TIME:20110421T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1302@cern.ch
DESCRIPTION:`Dyad' means a pair\, `dyadic' means binary\, and in this talk
I'll present some simple algorithms for rounding real vectors to 0-1 vect
ors without losing much. In particular\, we can round eigenvectors of real
matrices to 0-1 vectors with only a small log-factor loss. This leads to
interesting decompositions of real matrices into a small number of cut mat
rices -- each of which looks like a block of ones with zeroes all around (
up to some multiple) -- and can be thought of as rounding the singular val
ue decomposition (joint work with Ravindran Kannan and Nikhil Srivastava).
\n\nThe talk will be elementary and all are encouraged to attend.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1302
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1302
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum Query Complexity of Minor-closed Graph Properties
DTSTART;VALUE=DATE-TIME:20110427T103000Z
DTEND;VALUE=DATE-TIME:20110427T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1326@cern.ch
DESCRIPTION:We study the quantum query complexity of minor-closed graph pr
operties\, which include such problems as determining whether a graph is p
lanar\, is a forest\, or does not contain a path of a given length. We sho
w that most minor-closed properties---those that cannot be characterized b
y a finite set of forbidden subgraphs---have quantum query complexity Thet
a(n^{3/2}). To establish this\, we prove an adversary lower bound using a
detailed analysis of the structure of minor-closed properties with respect
to forbidden topological minors and forbidden subgraphs. On the other han
d\, we show that minor-closed properties (and more generally\, sparse grap
h properties) that can be characterized by finitely many forbidden subgrap
hs can be solved strictly faster\, in o(n^{3/2}) queries. Our algorithms a
re a novel application of the quantum walk search framework and give impro
ved upper bounds for several subgraph-finding problems.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=1326
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1326
END:VEVENT
BEGIN:VEVENT
SUMMARY:Continuum Percolation with Unbounded Radius Random Variable
DTSTART;VALUE=DATE-TIME:20110503T040000Z
DTEND;VALUE=DATE-TIME:20110503T050000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1340@cern.ch
DESCRIPTION:We show that in the Poisson continuum percolation model on the
plane there is a sharp transition from a regime which admits unbounded va
cant clusters to a regime which admits unbounded occupied clusters. This i
s joint work with Ghurumuruhan Ganesan.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=1340
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1340
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quadratic Programs and Some Constrained Stochastic Games
DTSTART;VALUE=DATE-TIME:20110512T053000Z
DTEND;VALUE=DATE-TIME:20110512T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1344@cern.ch
DESCRIPTION:We consider two player non-zero sum discounted cost single con
troller stochastic games. For such a player two controlled stochastic game
\, suppose\, player one has subscription type constraints and player two h
as realization based constraints that do not depend on first player's stra
tegies. We show that Nash equilibria of such constrained stochastic games
are in one-to-one correspondence with global minima of certain non-convex
quadratic programs. We show similar results for separable reward and state
independent stochastic games with subscription type constraints. Nash equ
ilibria for the above games can be computed in finite number of steps usin
g the available algorithms for computing global minima of non-convex QPs (
joint work with Vikas Vikram Singh).\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=1344
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1344
END:VEVENT
BEGIN:VEVENT
SUMMARY:Analysis of Voting Systems
DTSTART;VALUE=DATE-TIME:20110517T113000Z
DTEND;VALUE=DATE-TIME:20110517T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1382@cern.ch
DESCRIPTION:This talk will be an introduction to Social Choice Theory\, wh
ich analyses methods to combine preferences of large number of voters to o
btain a result that is "fair". We will define some reasonable conditions f
or fairness\, and prove the Arrows Impossibility theorem which states that
no voting system can satisfy all these conditions.\n\nReferences: This th
eory has been used to analyse an alternative voting scheme in UK\, where a
recent referendum about it was held\n(see http://www.youtube.com/watch?fe
ature=player_embedded&v=NS84zuf5_LQ#at=191).\n\nAmartya Sen's Nobel Lectur
e\n(http://nobelprize.org/nobel_prizes/economics/laureates/1998/sen-lectur
e.pdf)\n\nProof of Arrows Theorem\n(http://home.uchicago.edu/~preny/papers
/arrow-gibbard-satterthwaite.pdf)\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=1382
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1382
END:VEVENT
BEGIN:VEVENT
SUMMARY:Art Gallery Problem
DTSTART;VALUE=DATE-TIME:20110524T113000Z
DTEND;VALUE=DATE-TIME:20110524T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1391@cern.ch
DESCRIPTION:This talk will provide a brief introduction to the art gallery
problem\, which was first posed to Chvátal by Victor Klee in 1973. It ba
sically deals with finding out the minimum number of watchmen/guards neede
d to keep watch over an entire art gallery shaped like a simple polygon ha
ving n vertices. I will be stating some theorems/conjectures (alongwith pr
oof sketches) regarding the number of guards that should be sufficient and
necessary for guarding different types of simple polygons having n vertic
es. Then\, I will describe an elegant approximation algorithm for solving
this problem which was suggested by Ghosh in 1986\, that closely follows t
he greedy approach for the minimum set-cover problem.\n\nReferences: http:
//en.wikipedia.org/wiki/Art_gallery_problem\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=1391
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1391
END:VEVENT
BEGIN:VEVENT
SUMMARY:Counting\, Quantifiers and Algorithms
DTSTART;VALUE=DATE-TIME:20110526T060000Z
DTEND;VALUE=DATE-TIME:20110526T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1390@cern.ch
DESCRIPTION:Logic (first order or temporal) can be easily extended to desc
ribe counts\, parity and other numerical phenomena. The satisfiability pro
blem of the extended logic is easily seen to be undecidable\, even on mode
ls which are finite words. We examine situations in which the satisfiabili
ty problem on such models remains decidable (joint work with A V Sreejith)
.\n\nTemporal logic with modulo counting is Pspace-complete (the lower bou
nd holds without modulo counting\, by Sistla and Clarke). Two-variable fir
st order logic with modulo counting quantifiers is Expspace-complete (the
upper bound is by an exponential translation to temporal logic\, following
Etessami\, Vardi and Wilke). If positions can be checked for modulo congr
uence\, but parity is kept out of the picture\, temporal logic is complete
for the third level of the polynomial hierarchy Sigma3P\, while two-varia
ble first order logic is Nexptime-complete (the lower bound holds without
the congruence checking\, by Weis and Immerman). Finally\, in the special
case of a one-letter alphabet\, full first order logic with any number of
variables\, counting quantifiers and addition built-in is decidable (by Sc
hweikardt)\; in the case of modulo counting quantifiers we give a double e
xponential space upper bound. Without the modulo counting quantifiers this
is called Presburger arithmetic and is known to be complete for alternati
ng doubly exponential time with a linear number of alternations\, shown by
Berman. So a small gap remains.\n\nThe talk will concentrate on the algor
ithms rather than on details of complexity classes.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=1390
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1390
END:VEVENT
BEGIN:VEVENT
SUMMARY:NP-Hardness of Art Gallery Problem & Terrain Guarding Problem
DTSTART;VALUE=DATE-TIME:20110530T113000Z
DTEND;VALUE=DATE-TIME:20110530T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1410@cern.ch
DESCRIPTION:This will be a talk on the proof of NP hardness of the Art Gal
lery Problem (which was described by Pritam last week). Approximation algo
rithms for the terrain guarding problem will also be described.\n\nReferen
ces: http://www.tcs.tifr.res.in/~ghosh/artgallery-approx.pdf\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=1410
LOCATION:Colaba Campus A-212 (STCS Seminarf Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1410
END:VEVENT
BEGIN:VEVENT
SUMMARY:Random Walks and Electrical Networks : A Simple Approach
DTSTART;VALUE=DATE-TIME:20110607T113000Z
DTEND;VALUE=DATE-TIME:20110607T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1424@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
1424
LOCATION:Colaba Campus A-212 (STCS Seminar Room
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1424
END:VEVENT
BEGIN:VEVENT
SUMMARY:Real-Time High Fidelity Rigid Body Simulations for Virtual Environ
ments
DTSTART;VALUE=DATE-TIME:20110610T060000Z
DTEND;VALUE=DATE-TIME:20110610T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1323@cern.ch
DESCRIPTION:Rigid body simulation is an integral part of Virtual Environme
nts (VE) for autonomous planning\, training\, and design tasks. The underl
ying physics based simulation of VE must be accurate and computationally f
ast enough for the intended application\, which unfortunately are conflict
ing requirements. Two ways to perform fast and high fidelity physics based
simulation are: (1) model simplification\, and (2) parallel computation.
Model simplification can be used to allow simulation at an interactive rat
e while introducing an acceptable level of error. Currently\, manual model
simplification is the most popular way of performing simulation speed up
but it is time consuming. Hence\, in order to reduce the development time
of VEs\, automated model simplification is needed. I will present an autom
ated model simplification approach based on geometric reasoning\, spatial
decomposition\, and temporal coherence. Geometric reasoning is used to dev
elop an accessibility based algorithm for removing portions of geometric m
odels that do not play any role in rigid body to rigid body interaction si
mulation. Removing such inaccessible portions of the interacting rigid bod
y models has no influence on the simulation accuracy but reduces computati
on time significantly. Spatial decomposition is used to develop a clusteri
ng algorithm that reduces the number of fluid pressure computations result
ing in significant speedup of rigid body and fluid interaction simulation.
Temporal coherence algorithm reuses the computed force values from rigid
body to fluid interaction based on the coherence of fluid surrounding the
rigid body. The simulations can further be sped up by performing computing
on general purpose graphics processor (GP-GPU). Harnessing GP-GPU computi
ng technology requires development of parallel algorithms for the simulati
on. I will talk about the issues pertaining to the development of parallel
algorithms for rigid body simulations. The developed algorithms have enab
led real time high fidelity 6-DOF time domain simulation of Unmanned Sea S
urface Vehicles. The developed simulator can be used for autonomous motion
planning\, teleoperation\, and learning from demonstration.\n\nBiography:
Atul Thakur is a Ph.D. candidate in the Department of Mechanical Engineer
ing at the University of Maryland. His main research interests include mod
el simplification for physics based simulations for unmanned vehicles and
robot motion planning. Prior to joining University of Maryland in 2007\, h
e was working as Design Engineer at Aircraft Engine Performance Engineerin
g center of excellence at General Electric - India Technology Center\, Ban
galore. He received a Master of Technology (M. Tech.) degree in Manufactur
ing Engineering from the Indian Institute of Technology\, Bombay in 2006.
He received a Bachelor of Engineering (B.E.) degree in Production Engineer
ing from the University of Mumbai in 2003.\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=1323
LOCATION:Colaba Campus A-212
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1323
END:VEVENT
BEGIN:VEVENT
SUMMARY:Opportunistic Scheduling and Routing in Wireless Networks
DTSTART;VALUE=DATE-TIME:20110613T053000Z
DTEND;VALUE=DATE-TIME:20110613T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1411@cern.ch
DESCRIPTION:A centralized dynamic\, opportunistic scheduling decision has
to be made by a Base station (BS) to fair share the resources\, based on t
he current channel gains signaled by the mobiles. But the mobiles can be n
on-cooperative\, they may signal erroneously to improve their own utilitie
s. We first study the case of efficient scheduling (which gives least prio
rity to fairness) via Signaling Game. We find that this game admits only b
abbling equilibria at which the BS ignores the signals from mobiles. We th
en propose various approaches to enforce truthful signaling. We obtain a s
tochastic approximation based robust scheduling policy\, that combines est
imation and control. We further identify other equilibria that involve non
-truthful signaling. The scheduling policy that gives priority to fairness
can not be studied using a Signaling game. Nevertheless\, using general n
on-cooperative game theoretic framework\, we show that there exists only N
ash Equilibrium (NE)\, which resemble the babbling equilibria. However\, a
s the priority to fairness increases the existing fair schedulers become i
ncreasingly robust against non-cooperation. We then propose robust fair sc
hedulers which induce truth revealing NE as is done in the case of efficie
nt schedulers.\n\nIn wireless networks with limited connectivity\, data ro
uting becomes a challenging task. Users demanding data transfers can toler
ate delays and this fact is utilized to design effective routing solutions
\, utilizing the contact opportunities that arise because of the mobility
patterns of the elements of the network. These are called Delay Tolerant N
etworks (DTNs). In one such example\, message is spread across the contact
ed (relay) users and it spreads like epidemics till it reaches the destina
tion. The epidemics should be generated in a controlled manner so as to op
timize performance. Successful delivery probability of a message within a
given deadline is one such performance and it often takes the form of the
expectation of the exponent of some integral cost. We identify a general f
ramework from optimal control in finance\, known as risk sensitive control
\, which allows us to handle this multiplicative cost and obtain solutions
to several novel control problems in DTNs. New optimal control problems w
hich consider the effect of wireless propagation path loss factor and the
power constraints at the source and or the destination are proposed for DT
Ns. The possibility of non-threshold type optimal policies is established
for some of the control problems.\n\nAlternatively\, a BS is placed in a m
oving object (for example a Bus circulating in the area) to facilitate dat
a transfer. We study one such system\, ferry based wireless local area net
work (FWLAN)\, using the results of polling systems. Not much theory is av
ailable for the polling system that can model FWLANs as these need systems
with continuum of arrivals. We first obtain the stationary workload perfo
rmance of continuous polling systems via discretization approach and apply
the results to study FWLANs.\n\nBrief Biodata: Kavitha obtained her B.E.
degree in Electronics from UVCE\, Bangalore in 1994 and the M.Sc (Engg) an
d Ph.D. degrees respectively from the Departments of EE and ECE\, Indian I
nstitute of Science (IISc)\, Bangalore in 2002 and 2007. From 1994-2000\,
she was involved in the design and development of GPS\, CDMA and Voice ban
d modems at Accord Software and Systems\, Bangalore. She was an NBHM (Nati
onal Board for Higher Mathematics) post doctoral fellow at Tata Institute
of Fundamental Research (TIFR)\, Bangalore\, during 2007-08. From 2008 onw
ards\, she has been a post doctoral researcher with MAESTRO\, INRIA\, Soph
ia Antipolis\, France and LIA\, University of Avignon\, France.\n\nhttps:/
/indico.tifr.res.in/indico/conferenceDisplay.py?confId=1411
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1411
END:VEVENT
BEGIN:VEVENT
SUMMARY:Topics in MIMO Wireless Communication and Collaborative Estimation
DTSTART;VALUE=DATE-TIME:20110614T083000Z
DTEND;VALUE=DATE-TIME:20110614T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1426@cern.ch
DESCRIPTION:In this talk\, we discuss three problems. The first is on an a
synchronous multi-antenna wireless communication system and the later two
are on collaborative estimation.\n\na) It is known that we can exploit cer
tain type of asynchronism in a wireless communication system to our advant
age. We quantify the capacity of a multi-antenna communication system with
asynchronism\, and observe that it can show significant improvement over
a synchronous system.\n\nb) We analyze a local popularity based recommenda
tion algorithm that uses information about a set of "similar" users to rec
ommend an item to a user. For a particular random matrix model\, we would
identify regimes where the local popularity based algorithm works and wher
e it does not. We will also discuss some empirical results in light of our
theoretical results and compare with an approach based on low-rank matrix
completion.\n\nc) We consider a collection of regression experiments whic
h show clustered behavior. We study and compare several methods for this\,
demonstrate their applicability to the Yahoo Learning-to-rank Challenge (
YLRC) dataset\, and investigate an associated mathematical model. Based on
empirical evaluation on the YLRC dataset as well as simulated data\, we i
dentify that a local regression (LoR) scheme is a good practical choice: i
t yields best or near-best prediction performance at a reasonable computat
ional load\, and it is less sensitive to the choice of the algorithm param
eter. We will also discuss some analysis of the LoR method for an associat
ed mathematical model\, which sheds light on optimal parameter choice and
prediction performance.\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=1426
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1426
END:VEVENT
BEGIN:VEVENT
SUMMARY:Signal Processing and MIMO for Miltimode Fiber Links
DTSTART;VALUE=DATE-TIME:20110623T060000Z
DTEND;VALUE=DATE-TIME:20110623T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1446@cern.ch
DESCRIPTION:Multimode fibers (MMF) are often used in short haul optical ne
tworks owing to favourable geometries and inexpensive packaging costs. How
ever\, the performance of multimode fibers has been limited primarily by t
he presence of high modal dispersion owing to large core diameters. While
electronic dispersion compensation methods improve the bandwidth-distance
product of MMF\, they do not utlilize the fundamental diversity present in
the different modes of the multimode fiber. In this work\, we draw from d
evelopments in communication theory and signal processing for wireless com
munication in order to motivate the use of multiple-input multiple-output
signal processing concepts to enable high-data rates through MMFs. We deve
lop theoretical models and propose signaling algorithms to improve data ra
tes with multiple modulators and detectors\, and experimentally verify the
validity of the theory on a multimode fiber link. We further explore the
applicability of this theory to other multimode waveguides and consider th
e design of specific electro-optic devices to maximize data rates in conju
nction with signal processing.\n\nBio: Kumar Appaiah is a graduate student
of Electrical and Computer Engineering at The University of Texas at Aust
in\, USA. His research interests are primarily in communication theory and
applications of signal processing for wireless and optical links. Prior t
o joining UT Austin\, he earned a B.Tech. in Electrical Engineering and M.
Tech. in Communication Engineering from IIT Madras.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=1446
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1446
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Concert Queueing Game with a Finite Homogeneous Population
DTSTART;VALUE=DATE-TIME:20110704T103000Z
DTEND;VALUE=DATE-TIME:20110704T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1452@cern.ch
DESCRIPTION:We consider the non-cooperative choice of arrival times by ind
ividual users\, or customers\, to a service system that opens at a given t
ime\, and where users queue up and are served in order of arrival. Each us
er wishes to obtain service as early as possible\, while minimizing the ex
pected wait in the queue. This problem was recently studied within a simpl
ified fluid model. Here we address the non-asymptotic stochastic system\,
assuming a finite (possibly random) number of homogeneous users\, exponent
ial service times\, and linear cost functions. In this setting we show tha
t there exists a unique Nash equilibrium\, which is symmetric across users
\, and characterize the equilibrium arrival-time distribution of each user
in terms of a corresponding set of differential equations. We further est
ablish convergence of the Nash equilibrium solution to that of the associa
ted fluid model as the number of users increases to infinity. We also show
that the price of anarchy in our system exceeds 2\, but approaches this v
alue for a large population size.\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=1452
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1452
END:VEVENT
BEGIN:VEVENT
SUMMARY:Convex Optimization and Variational Inequalities
DTSTART;VALUE=DATE-TIME:20110704T043000Z
DTEND;VALUE=DATE-TIME:20110704T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1466@cern.ch
DESCRIPTION:This talk concerns the basic facts of convex optimization and
variational inequalities. We will discuss about subgradients and provide i
mportant examples. Further we will discuss some important algorithms for n
on-smooth convex optimization like the projected subgradient method. We wi
ll also discuss some important algorithms for solving finite dimensional v
aritaional inequality problems.\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=1466
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1466
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dual Extrapolation Method for Solving Variational Inequalities: Th
e Approach of Nesterov
DTSTART;VALUE=DATE-TIME:20110704T060000Z
DTEND;VALUE=DATE-TIME:20110704T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1467@cern.ch
DESCRIPTION:In this talk we essentially discuss a work on variational ineq
ualities by Yuri Nestrov where the so called prox-functions play a major r
ole. We discuss the approach in details giving complexity bounds and relat
ed results.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=1467
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1467
END:VEVENT
BEGIN:VEVENT
SUMMARY:Physics of Computation
DTSTART;VALUE=DATE-TIME:20110721T103000Z
DTEND;VALUE=DATE-TIME:20110721T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1517@cern.ch
DESCRIPTION:Our daily experiences with life and silicon constantly remind
us that computation requires energy. Prima facie\, an attractive analogy w
ith steam engines suggests itself -- perhaps computers are machines to con
vert energy into "computational work". On the other hand\, careful analysi
s by Charles Bennett has led him to suggest that computation can be done f
or arbitrarily little energy per step. I will describe the first steps of
a research program that seeks to reconcile this apparent conflict\, in the
hope that such analysis will some day aid the study of biological process
es\, where computation is ubiquitous and Darwinian evolution provides grea
t incentive to be efficient.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=1517
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1517
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stochastic Games
DTSTART;VALUE=DATE-TIME:20110729T103000Z
DTEND;VALUE=DATE-TIME:20110729T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1539@cern.ch
DESCRIPTION:I want to present a portion of Shapley's paper introducing sto
chastic games. We'll see what the notion of value is\, for such types of g
ames. We'll prove that every stochastic game has a value and that there is
always a pair of stationary strategies which are optimal. Please find the
paper attached.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=1539
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1539
END:VEVENT
BEGIN:VEVENT
SUMMARY:Constructive Aspects of the Lovasz Local Lemma and their Applicati
ons
DTSTART;VALUE=DATE-TIME:20110802T103000Z
DTEND;VALUE=DATE-TIME:20110802T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1532@cern.ch
DESCRIPTION:Recent years have seen significant progress on the algorithmic
aspects of the Lovasz Local Lemma: e.g.\, one can now handle super-polyno
mially many events that need to be avoided. I will survey this general are
a\, as well as my joint work with Bernhard Haeupler and Barna Saha.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1532
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1532
END:VEVENT
BEGIN:VEVENT
SUMMARY:Constructing Some Finite Normal Form Games
DTSTART;VALUE=DATE-TIME:20110803T103000Z
DTEND;VALUE=DATE-TIME:20110803T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1537@cern.ch
DESCRIPTION:We present approaches for constructing finite normal form game
s that ensure constructed games have desirable outcomes.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=1537
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1537
END:VEVENT
BEGIN:VEVENT
SUMMARY:Random Regular Graphs: Generation and Cover Time
DTSTART;VALUE=DATE-TIME:20110803T050000Z
DTEND;VALUE=DATE-TIME:20110803T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1548@cern.ch
DESCRIPTION:A graph is $r$-regular if all its vertices have the same degre
e $r$. A random $r$-regular graph with $n$ vertices is a graph sampled uni
formly at random from the set of all $r$-regular graphs on $n$ vertices. T
here are various sampling methods for this. In this talk we look at one of
them: The Configuration Model. We then look at the cover time of a random
walk on such a graph\, which roughly speaking\, is the time taken (in exp
ectation) by the walk to visit all the vertices. This cover time calculati
on\, due to Colin Cooper and Alan Frieze\, uses generating function techni
que and results from complex analysis\, which we illustrate in this talk.\
n\n(Key words.) Random regular graphs\, configuration model\, cover time\,
combinatorics.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=1548
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1548
END:VEVENT
BEGIN:VEVENT
SUMMARY:Recognizing Point Visibility Graphs
DTSTART;VALUE=DATE-TIME:20110804T103000Z
DTEND;VALUE=DATE-TIME:20110804T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1531@cern.ch
DESCRIPTION:We explore the visibility graph of a point set. We list some c
lassical works on configurations of points and straight lines on the plain
and then proceed to deduce some combinatorial properties of point visibil
ity graphs. After that we identify some necessary conditions for a graph f
or being a visibility graph of some point set and discuss algorithms\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1531
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1531
END:VEVENT
BEGIN:VEVENT
SUMMARY:What Risks Lead to Ruin
DTSTART;VALUE=DATE-TIME:20110809T103000Z
DTEND;VALUE=DATE-TIME:20110809T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1557@cern.ch
DESCRIPTION:Insurance transfers losses associated with risks to the insure
r for a price\, the premium. We adopt the collective risk approach. Namely
\, we abstract the problem to include just two agents: the insured and the
insurer. We are interested in scenarios where the underlying model for th
e loss distribution is not very well known\, and the potential losses can
also be quite high\, which is of potential interest\, for instance\, in in
suring against loss-of-use risk for services on offer over the Internet\,
where models for the statistics of the loss are not well established. It i
s then natural to adopt a nonparametric formulation. Considering a natural
probabilistic framework for the insurance problem\, assuming independent
and identically distributed (i.i.d.) losses\, we derive a necessary and su
fficient condition on nonparametric loss models such that the insurer rema
ins solvent despite the losses taken on.\n\nIn more detail\, we model the
loss at each time by a nonnegative integer. An insurer’s scheme is defin
ed by the premium demanded by the insurer from the insured at each time as
a function of the loss sequence observed up to that time. The insurer is
allowed to wait for some period before beginning to insure the process\, b
ut once insurance commences\, the insurer is committed to continue insurin
g the process. All that the insurer knows is that the loss sequence is a r
ealization from some i.i.d. process with marginal law in some set $P$ of p
robability distributions on the nonnegative integers. The insurer does not
know which $p \\in{\\cal P}$ describes the distribution of the loss seque
nce. The insurer goes bankrupt when the loss incurred exceeds the built up
buffer of reserves from premiums charged so far.\n\nWe show that a finite
support nonparametric loss model of this type is insurable if it contains
no “deceptive” distributions. Here the notion of ``deceptive” distr
ibution is precisely defined in information-theoretic terms. Note that\, e
ven though we assume a finite support for each $p \\in{\\cal P}$\, there i
s no absolute bound assumed on the possible loss at any time. The necessar
y background from information theory and risk theory as well as some motiv
ation for the problem formulation will be provided during the talk (joint
work with Narayana Prasad Santhanam (University of Hawaii\, Manoa)).\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1557
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1557
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dynamical Consequences of Bandpass Feedback in a Biomolecular Sign
aling Circuit
DTSTART;VALUE=DATE-TIME:20110811T041500Z
DTEND;VALUE=DATE-TIME:20110811T051500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1582@cern.ch
DESCRIPTION:Diverse engineering functions can be generated when simple mod
ules are connected in different configurations. Similar modules also exist
in biology\, for example\, within circuits of interacting molecules that
map to diverse cellular behavior. Understanding this mapping is both a fun
damental problem in biology and also challenging from a mathematical persp
ective. In this talk\, I will present our work on how cellular transitions
from one state to another in a canonical system are encoded molecularly.\
n\nUnder conditions of nutrient limitation\, cells of the bacteria Bacillu
s subtilis transition from a state of growth to a state of dormancy. Progr
ession to dormancy is controlled by a biomolecular signaling circuit with
multiple feedback loops. These regulatory interactions are "bandpass"-like
\, in the sense that the output is active in a limited band of input value
s\, and have recently been shown to pulse in a periodic fashion. However\,
the impact of these pulsed bandpass interactions on circuit dynamics prec
eding dormancy remains unclear. In order to address this question\, we est
imated key features of the bandpass interactions at the single-cell level
and analyzed them in the context of a simple mathematical model. The model
predicted the emergence of a delayed phase shift in the activities of the
circuit components\, as well as the existence of a stable state\, interme
diate between growth and dormancy\, embedded within the dynamical structur
e of the circuit. To test the model\, we used time-lapse fluorescence micr
oscopy to measure dynamics of single cells entering dormancy. We observed
the delayed phase shift emerging during the progression\, while a re-engin
eering of the sporulation circuit revealed behavior resembling the predict
ed additional state. These results show that periodically-driven bandpass
feedback loops can give rise to complex dynamics in this transition from g
rowth to dormancy.\n\nIn the context of these results\, I will discuss res
earch challenges in exploring the functional criticality of such circuit d
ynamics\, the role of feedback in related circuit configurations\, and whe
ther these design insights can be used in other engineering processes.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1582
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1582
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Upper Bound on the Bisection Width of d-Regular Graphs
DTSTART;VALUE=DATE-TIME:20110812T103000Z
DTEND;VALUE=DATE-TIME:20110812T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1593@cern.ch
DESCRIPTION:The 'bisection width' of a graph is defined as the minumum num
ber of edges that need to be removed to partition the graph into two equal
halves. If the graph is 'sparse'\, then we don't expect this quantity to
be large. In this talk\, I will present a result of Noga Alon\, which give
s an upper bound on the bisection width of any d-regular graph\, where the
number of vertices is 'much more' than d. Time permitting\, I will also d
iscuss some other interesting cut problems and what we know about them.\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1593
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1593
END:VEVENT
BEGIN:VEVENT
SUMMARY:Near-Popular Matchings in the Roommates Problem
DTSTART;VALUE=DATE-TIME:20110926T113000Z
DTEND;VALUE=DATE-TIME:20110926T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1712@cern.ch
DESCRIPTION:Our input is a graph G = (V\, E) where each vertex ranks its n
eighbors in a strict order of preference. The problem is to compute a matc
hing in G that captures the preferences of the vertices in a popular way.
Matching M is more popular than matching M' if the number of vertices that
prefer M to M' is more than those that prefer M' to M. The unpopularity f
actor of M measures by what factor any matching can be more popular than M
. We show that G always admits a matching whose unpopularity factor is O(l
og|V|)\, and such a matching can be computed in linear time. In our proble
m the optimal matching would be a least unpopularity factor matching. We w
ill show that computing such a matching is NP-hard. In fact\, for any $eps
ilon$\, it is NP-hard to compute a matching whose unpopularity factor is a
t most $4/3 - epsilon$ of the optimal.\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=1712
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1712
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lookup Tables\, Suffix Trees and Suffix Arrays (Part-I)
DTSTART;VALUE=DATE-TIME:20111010T103000Z
DTEND;VALUE=DATE-TIME:20111010T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1729@cern.ch
DESCRIPTION:Considering the large volume of sequence data that many comput
ational biology applications must deal with\, proper organization of the d
ata to facilitate fast access is important to achieve desirable run-times.
From this perspective\, string data structures serve the same purpose for
biological sequence data as binary search trees serve for ordered numeric
data\, and quadtrees serve for spatial data.\n\nString data structures ar
e ideal for uncovering exact matching patterns in sequences. They are also
useful when performing approximate matches where only a small number of d
ifferences are permitted.\n\nWe will try to provide a detailed introductio
n to the three most frequently used string data structures in computationa
l molecular biology — lookup tables\, suffix trees and suffix arrays. Th
e focus will be on algorithms for constructing these data structures. We w
ill also explore the relationships between these data structures and provi
de several illustrations of biological applications where suffix trees pla
y a central role.\n\nReferences:\nhttp://en.wikipedia.org/wiki/Suffix_tree
\nhttp://en.wikipedia.org/wiki/Suffix_array\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=1729
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1729
END:VEVENT
BEGIN:VEVENT
SUMMARY:Beyond Moore's Law Technologies and Architectures
DTSTART;VALUE=DATE-TIME:20111020T060000Z
DTEND;VALUE=DATE-TIME:20111020T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1733@cern.ch
DESCRIPTION:Conventional shrinking methods to improve VLSI chip performanc
e by continual scaling of device and interconnect geometries may allow CMO
S juggernaut to reach about 22 nm nodes. During the post-shrinking era\, a
host of nanoscale technologies such as quantum tunneling devices\, plasmo
n based transistors\, ionic transport based crossbar structures\, carbon n
ano-tube s\, grapheme s\, self- assembled array of quantum dots\, spin-pol
arized magnetic tunneling junction devices\, and molecular transistors are
likely to emerge as commercially viable technologies that will sustain th
e demands for exponential economic growth throughout the present millenniu
m.\n\nQuantum tunneling in nanometric devices augurs a revolutionary shift
of paradigm for circuit and CAD tools design that must account for quantu
m effects as well as local interactions between self- assembled circuit el
ements. These circuit elements may consist of a 2-dimensional array of sel
f-organized quantum dots that can be instrumented to perform cellular auto
mata class of algorithms or a 3-dimensional array of self-organized nanowi
res to perform a random Boolean network (RBN) class of algorithms. The tal
k will briefly introduce several Boolean and neuromorphic nanoarchitecture
s consisting of 2-D array of amorphous-Silicon based memristor devices\, s
tress-assisted spin polarized nanomagnets\, and surface plasmon polariton
(SPP) based THz active and passive devices.\n\nBiography: Professor Pinaki
Mazumder received his PhD in Electrical and Computer Engineering from the
University of Illinois at Urbana-Champaign in 1988. Prior to that\, he re
ceived his MS degree in Computer Science from University of Alberta in Can
ada\, BS degree in Electrical Engineering from Indian Institute of Science
at Bangalore\, and BSc Physics Honors degree from Guwahati University in
India. Currently\, he is a Professor of Electrical Engineering and Compute
r Science at the University of Michigan where he has been teaching for the
past 24 years. He spent three years at National Science Foundation servin
g as the lead Program Director of Emerging Technologies Program in the CIS
E Directorate as well as leading the Quantum\, Molecular\, and High Perfor
mance Simulation Program in the Engineering Directorate. He had worked for
six years in industrial R&D laboratories which included AT&T Bell Laborat
ories\, where he started the first C language based modeling techniques fo
r VLSI synthesis tool in 1985. Professor Mazumder spent his sabbatical at
Stanford University\, University of California at Berkeley\, and NTT Cente
r Research Laboratory in Japan. He has published over 250 technical papers
and 4 books on various aspects of VLSI technology and systems. His resear
ch interest includes CMOS VLSI design\, semiconductor memory systems\, CAD
tools and circuit designs for emerging technologies including quantum MOS
\, spintronics\, plasmonics\, and resonant tunneling devices. Prof. Mazumd
er was a recipient of Digital's Incentives for Excellence Award\, BF Goodr
ich National Collegiate Invention Award\, DARPA Research Excellence Award\
, and IEEE Distinguished Lecturer. Prof. Mazumder is an AAAS Fellow (2007)
and an IEEE Fellow (1999) for his distinguished contributions to the fiel
d of VLSI.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=1733
LOCATION:Colaba Campus A-269 (DAA Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1733
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lookup Tables\, Suffix Trees and Suffix Arrays (Part-II)
DTSTART;VALUE=DATE-TIME:20111024T103000Z
DTEND;VALUE=DATE-TIME:20111024T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1744@cern.ch
DESCRIPTION:Considering the large volume of sequence data that many comput
ational biology applications must deal with\, proper organization of the d
ata to facilitate fast access is important to achieve desirable run-times.
From this perspective\, string data structures serve the same purpose for
biological sequence data as binary search trees serve for ordered numeric
data\, and quadtrees serve for spatial data.\n\nString data structures ar
e ideal for uncovering exact matching patterns in sequences. They are also
useful when performing approximate matches where only a small number of d
ifferences are permitted.\n\nWe will try to provide a detailed introductio
n to the three most frequently used string data structures in computationa
l molecular biology — lookup tables\, suffix trees and suffix arrays. Th
e focus will be on algorithms for constructing these data structures. We w
ill also explore the relationships between these data structures and provi
de several illustrations of biological applications where suffix trees pla
y a central role.\n\nReferences:\n\nhttp://en.wikipedia.org/wiki/Suffix_tr
ee\nhttp://en.wikipedia.org/wiki/Suffix_array\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=1744
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1744
END:VEVENT
BEGIN:VEVENT
SUMMARY:Few Surprizes From Combinatorics And Number Theory
DTSTART;VALUE=DATE-TIME:20111031T103000Z
DTEND;VALUE=DATE-TIME:20111031T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1755@cern.ch
DESCRIPTION:Consider a natural number n. Let g denote the number of ones i
n the binary representation of n and h denote the greatest number such tha
t two raised to h divides n!. Then n=h+g. In the talk we will prove this a
nd few other "surprizing" results.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=1755
LOCATION:Colaba Campus A-212 (STCS Seminar Room
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1755
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximate Colouring using Semi Definite Programming
DTSTART;VALUE=DATE-TIME:20111114T103000Z
DTEND;VALUE=DATE-TIME:20111114T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1770@cern.ch
DESCRIPTION:Determining the chromatic number of a graph is known to be NP-
hard. We will consider the problem of coloring a k-colorable graph on n ve
rtices with n^{alpha} colors (where alpha \n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=1770
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1770
END:VEVENT
BEGIN:VEVENT
SUMMARY:Minimum and Maximum against k Lies
DTSTART;VALUE=DATE-TIME:20111121T103000Z
DTEND;VALUE=DATE-TIME:20111121T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1789@cern.ch
DESCRIPTION:We consider an n-element set X with an unknown total ordering.
The ordering can be accessed via an oracle that\, given two elements x\,y
in X\, tells us whether x < y or x > y. It is easy to see that the minimu
m element of X can be found using(n-1) comparisons. One of the nice little
surprises in computer science is that if we want to find both the minimum
and the maximum\, we can do significantly better than finding the minimum
and the maximum separately. A neat 1972 result of Pohl asserts that (ceil
(3n/2)-2) comparisons are sufficient\, and also necessary in the worst cas
e\, for finding both the minimum and the maximum of an n-element totally o
rdered set. Here we consider the problem of determining both the minimum a
nd the maximum in the case where the oracle is not completely reliable. It
may sometimes give a false answer\, but only at most k times during the w
hole computation\, where k is a given parameter. We refer to this model as
computation against k lies. Let us stress that we admit repeating the sam
e query to the oracle several times\, and each false answer counts as a li
e. If repeated queries were not allowed\, or if the oracle could always gi
ve the wrong answer to a particular query\, then the minimum can never be
determined accurately. So\, for example\, if we repeat a given query (2k +
1) times\, we always get the correct answer by majority vote. The problem
of finding both the minimum and the maximum against k lies was investigat
ed by Aigner\, who proved in 1997 that (k + O(k0.5))n comparisons always s
uffice. In this talk\, we will show an improvement on this by providing an
algorithm that requires at most (k+1+C)n+O(k3) comparisons for some const
ant C. The best known lower bounds for the number of comparisons necessary
to determine both the minimum and the maximum against k lies have the for
m (k+1+c_{k})n-D\, where D is a small constant and the c_{k} are as follow
s : c_{0} = 1/2 [Pohl]\, c_{1} = 23/32\, and c_{k} = Omega(2^{-5k/4}) as k
goes to infinity.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=1789
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1789
END:VEVENT
BEGIN:VEVENT
SUMMARY:Spectral Clustering and Transductive Learning for Graphical Data
DTSTART;VALUE=DATE-TIME:20111128T103000Z
DTEND;VALUE=DATE-TIME:20111128T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1810@cern.ch
DESCRIPTION:Suppose we want to cluster scientists into different groups su
ch that in each group scientists have some research interest in common. Th
e data available to us is the coauthor relationships. How do we classify u
sing spectral clustering ? Now suppose some of the scientists are already
labeled. How do we use that information effectively? In addition\, we have
the data regarding the political viewpoints of the scientists. Now how do
we combine both sets of data to get a useful clustering ? We will look in
to an algorithm proposed by Zhous et. al.\n\nReferences:\n\n1) Learning fr
om Labeled and Unlabeled Data on a Directed Graph\, Zhou et.al.\, 2005\n\n
2) Spectral Clustering and Transductive Learning with Multiple Views\, Zho
u et. al.\, 2007\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=1810
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1810
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximation Algorithm for Security Games with Costly Resources
DTSTART;VALUE=DATE-TIME:20111129T103000Z
DTEND;VALUE=DATE-TIME:20111129T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1754@cern.ch
DESCRIPTION:In recent years\, algorithms for computing game-theoretic solu
tions have been developed for real-world security domains. These games are
between a defender\, who must allocate her resources to defend potential
targets\, and an attacker\, who chooses a target to attack. Existing work
has assumed the set of defender’s resources to be ﬁxed. This assumptio
n precludes the effective use of approximation algorithms\, since a slight
change in the defender’s allocation strategy can result in a massive ch
ange in her utility. In contrast\, we consider a model where resources are
obtained at a cost\, initiating the study of the following optimization p
roblem: Minimize the total cost of the purchased resources\, given that ev
ery target has to be defended with at least a certain probability. We give
an efficient logarithmic approximation algorithm for this problem (joint
work with Vincent Conitzer and Kamesh Munagala).\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=1754
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1754
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Some Special Cases of the Entropy Photon-Number Inequality
DTSTART;VALUE=DATE-TIME:20111130T103000Z
DTEND;VALUE=DATE-TIME:20111130T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1790@cern.ch
DESCRIPTION:One of the aims of information theory is to find out the capac
ity of noisy channels defined as the ultimate limit on communication rates
. If an optical communication link is modeled quantum mechanically\, then
the task of finding the capacity (for sending classical data) relies on ce
rtain minimum output entropy conjecture which in turn is implied by the en
tropy photon number inequality (EPnI) conjecture. It has been a longstandi
ng problem in quantum information theory to resolve these conjectures. We
show that the EPnI holds where one of the input states is the vacuum state
and for several candidates of the other input state that includes the cas
es when the state has the eigenvectors as the number states and either has
only two non-zero eigenvalues or has arbitrary number of non-zero eigenva
lues but is a high entropy state. We also discuss the conditions\, which i
f satisfied\, would lead to an extension of these results (joint work with
Smarajit Das and Siddharth Muthukrishnan).\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=1790
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1790
END:VEVENT
BEGIN:VEVENT
SUMMARY:Towards Coding for Maximum Errors in Interactive Communication
DTSTART;VALUE=DATE-TIME:20111130T083000Z
DTEND;VALUE=DATE-TIME:20111130T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1822@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
1822
LOCATION:Colaba Campus
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1822
END:VEVENT
BEGIN:VEVENT
SUMMARY:Towards Coding for Maximum Errors in Interactive Communication
DTSTART;VALUE=DATE-TIME:20111130T083000Z
DTEND;VALUE=DATE-TIME:20111130T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1823@cern.ch
DESCRIPTION:We show that it is possible to encode any communication protoc
ol between two parties so that the protocol succeeds even if a (1/4 − ep
silon) fraction of all symbols transmitted by the parties are corrupted ad
versarially\, at a cost of increasing the communication in the protocol by
a constant factor (the constant depends on epsilon). This encoding uses a
constant sized alphabet. This improves on an earlier result of Schulman\,
who showed how to recover when the fraction of errors is bounded by 1/240
. We also show how to simulate an arbitrary protocol with a protocol using
the binary alphabet\, a constant factor increase in communication and tol
erating a 1/8 − epsilon fraction of errors.\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=1823
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1823
END:VEVENT
BEGIN:VEVENT
SUMMARY:Synchrony Induced Brain-like Computing in Artificial Molecular Sys
tems
DTSTART;VALUE=DATE-TIME:20111205T053000Z
DTEND;VALUE=DATE-TIME:20111205T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1786@cern.ch
DESCRIPTION:Nearly 3.5 billion years back microtubule came into being\, wh
ile unravelling its electronics we found its synchrony induced behavior (1
)\, which enables it to exhibit remarkable electronic properties never see
n before. Our results show that synchronous behavior leads to non-linear f
requency pulling which enables neutral network to solve problems much fast
er. Inspired by these process we devised organic monolayer\, and solved tw
o intractable problems on this surface in finite time (2). We are now buil
ding artificial molecular systems we call it nano-brain for robots that wi
ll learn & take decisions by itself without human intervention. We do not
need to write program or algorithm\, it is a new simple way to design mole
cular structure to execute around hundred "if-then" arguments simultaneous
ly\, note that simultaneity is not parallel. Thus\, we combine results obt
ained from electronics studies of real bio-materials\, devise computing pr
inciples (3) and then implement it into building artificial systems\, in a
ll our works we suggest that the new generation computing would be communi
cating "one-to-many at-a-time".\n\n(1) Radiofrequency induced ultra-fast a
ssembly of microtubules and their length-independent electronic properties
\, Satyajit Sahu\, Subrata Ghosh\, Kazuto Hirata\, Daisuke Fujita\, Anirba
n Bandyopadhyay\, Nature Materials (2011)\, in press.\n\n(2) Massively par
allel computing on an organic molecular layer\, Anirban Bandyopadhyay\, Ra
njit Pati\, Satyajit Sahu\, Ferdinand Peper\, Daisuke Fujita\, Nature Phys
ics (2010)\n\n(3) A 16-bit parallel processing in a molecular assembly\, A
nirban Bandyopadhyay\, Somobrata Acharya\, Proc. Natl. Acad. Sci. (2008).\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1786
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1786
END:VEVENT
BEGIN:VEVENT
SUMMARY:Online Matching and Ad Allocation
DTSTART;VALUE=DATE-TIME:20111207T090000Z
DTEND;VALUE=DATE-TIME:20111207T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1835@cern.ch
DESCRIPTION:The spectacular success of search and display advertising and
its huge growth potential has attracted the attention of researchers from
many aspects of computer science. A core problem in this area is that of A
d allocation\, an inherently algorithmic and game theoretic question - tha
t of matching ad slots to advertisers\, online\, under demand and supply c
onstraints. Put very simply: the better the matching\, the more efficient
the market.\n\nThe seminal algorithmic work on online matching\, by Karp\,
Vazirani and Vazirani\, was done over two decades ago\, well before this
motivation existed. In this talk\, I will present an overview of several k
ey algorithmic papers in this area\, starting with its purely academic beg
innings\, to the general problem motivated by the application. The theory
behind these algorithms involves new combinatorial\, probabilistic and lin
ear programming techniques. Besides the analytical results\, I will also t
ouch upon how these algorithmic ideas can be applied in practice.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1835
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1835
END:VEVENT
BEGIN:VEVENT
SUMMARY:Depth-Independent Lower Bounds on the Communication Complexity of
Read-Once Boolean Formulas
DTSTART;VALUE=DATE-TIME:20111210T053000Z
DTEND;VALUE=DATE-TIME:20111210T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1855@cern.ch
DESCRIPTION:We show lower bounds of $Omega(sqrt{n})$ on the randomized com
munication complexity\, respectively\, of all $n$-variable read-once Boole
an formulas. Our results complement the recent lower bound of $Omega(n/8^d
)$ by Leonardos and Saks and $Omega(n/2^{Omega(dlog d)})$ by Jayram\, Kopp
arty and Raghavendra for randomized communication complexity of read-once
Boolean formulas with depth $d$. We obtain our result by "embedding" eithe
r the Disjointness problem or its complement in any given read-once Boolea
n formula.\n\nIt is a small paper. I will try to present some additional r
esults proofs of which is not included in this paper. I will need 40 mins
to complete the presentation.\n\n(Author: Rahul Jain\, Hartmut Klauck\, Sh
engyu Zhang)\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=1855
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1855
END:VEVENT
BEGIN:VEVENT
SUMMARY:Generalized Network Sharing bound and Two-Unicast Networks
DTSTART;VALUE=DATE-TIME:20111214T083000Z
DTEND;VALUE=DATE-TIME:20111214T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1803@cern.ch
DESCRIPTION:The talk will be in two parts.\n\nIn the first part\, we consi
der the two-unicast problem in wireline networks\, i.e. the problem of com
munication over a network with two sources and two destinations\, each sou
rce with a message for its own destination. Our interest is in investigati
ng the network coding capacity region for this problem. We develop a new o
uter bound that is a simple improvement over an existing bound in the lite
rature called the Network Sharing bound [Yan\, Yang\, Zhang]. We call our
bound the Generalized Network Sharing (GNS) bound. We discover some intere
sting properties of this bound with regard to two-unicast networks (oint w
ork with Prof. David Tse and Prof. Venkat Anantharam).\n\nIn the second pa
rt\, we consider two-unicast in linear deterministic networks. The linear
deterministic model has been very successful in characterizing approximate
ly capacity regions of Gaussian networks. By developing a GNS bound for la
yered linear deterministic networks\, we find an interesting analogue of a
result obtained for two-unicast wireline networks by Chih-Chun Wang and N
ess Shroff. Further\, by providing achievable schemes and matching outer b
ounds\, we completely characterize the capacity region of a class of two-u
nicast layered linear deterministic networks (joint work with I-Hsiang Wan
g and Prof. David Tse).\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=1803
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1803
END:VEVENT
BEGIN:VEVENT
SUMMARY:Observing Markov Chains by Timed Automata
DTSTART;VALUE=DATE-TIME:20111216T060000Z
DTEND;VALUE=DATE-TIME:20111216T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1863@cern.ch
DESCRIPTION:Markov chains are omnipresent. They are used as semantical bac
kbone of Markovian queueing networks\, stochastic Petri nets\, stochastic
process algebras\, and calculi for systems biology. Abundant applications
in performance\, reliability and dependability analysis use Markov chains.
The specification of measures-of-interest is however quite limited\, and
dedicated algorithms are typically developed for every new measure.\n\nIn
this talk\, we will survey model checking techniques for Markov chains. On
the one hand\, temporal logics allow to express most measures of interest
as well as measures that go beyond the classical one. On the other hand\,
a single model-checking algorithm suffices to deal with all measures that
can be expressed.\n\nWe will describe the basic model-checking algorithms
\, covering both discrete-time and continuous-time Markov chains\, and rep
ort on recent progress on using timed automata as requirements (rather tha
n temporal logic).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=1863
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1863
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Logic of Interactive Proofs
DTSTART;VALUE=DATE-TIME:20111219T060000Z
DTEND;VALUE=DATE-TIME:20111219T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1793@cern.ch
DESCRIPTION:We propose a logic of interactive proofs as the first and main
step towards an intuitionistic foundation for interactive computation to
be obtained via an interactive analog of the Godel Kolmogorov-Artemov defi
nition of intuitionistic logic as embedded into a classical modal logic of
proofs\, and of the Curry-Howard isomorphism between intuitionistic proof
s and typed programs. Our interactive proofs effectuate a persistent epist
emic impact in their intended communities of peer reviewers that consists
in the induction of the (propositional) knowledge of their proof goal by m
eans of the (individual) knowledge of the proof with the interpreting revi
ewer. That is\, interactive proofs effectuate a transfer of propositional
knowledge -- (to-be-)known facts -- via the transfer of certain individual
knowledge -- (to-be-)known proofs -- in multi-agent systems. In other wor
ds\, we as a community have the formal common knowledge that a proof is th
at which if known to one of our peer members would induce the knowledge of
its proof goal with that member.\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=1793
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1793
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Role of Knowledge in Games
DTSTART;VALUE=DATE-TIME:20120102T050000Z
DTEND;VALUE=DATE-TIME:20120102T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1892@cern.ch
DESCRIPTION:What agents do depends not only on what they desire but also o
n what they believe and what they believe others will do. This makes it po
ssible to influence agents by revealing or withholding information. We wil
l give a short account of the Kripke structure model of knowledge\, how it
is affected by information exchange and how a knowledge manipulator can i
nfluence the actions of other agents by sending signals.\n\nAs an aside we
mention the role of temperament. What an agent will do in the presence of
uncertainty will depend not only on the knowledge and preferences of the
agent\, but also on the agent's temperament - whether cautious or aggressi
ve. We present some results about how an agent A's knowledge of another ag
ent B's temperament can help A to predict B's choices\, and thus allow A t
o make better choices herself.\n\nOne application of knowledge manipulatio
n is campaigning where a politician running for office decides what to say
about various issues with a view to affecting the attitudes of the voters
towards himself.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.p
y?confId=1892
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1892
END:VEVENT
BEGIN:VEVENT
SUMMARY:Welfare and Profit Maximization with Procurement Costs
DTSTART;VALUE=DATE-TIME:20120103T090000Z
DTEND;VALUE=DATE-TIME:20120103T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1809@cern.ch
DESCRIPTION:What is the problem?\n\nA limited resource needs to be allocat
ed amongst a set of self-interested agents. For instance\, a company manag
er wants to allocate resources such as labour force or technical infrastru
cture to various projects of the company. How should the manager allocate
the resources to maximize the output of the company? The manager faces two
issues that might hinder her in achieving the objective. The first is tha
t the resources are limited\, hence she needs to choose whom to allocate a
nd whom not to. The second is that the manager is not aware of the needs o
f various projects and each project team is self-interested\, i.e.\, each
project team is interested only in furthering its own project and might li
e about the value of the project or the resources they need.\n\nI like the
problem. So\, what is your contribution? Allocating limited resources in
a game-theoretic setting has been considered previously in research litera
ture. The main contribution of our work is that we model limitation of res
ources in a more realistic manner. Previous literature has modeled the lim
ited nature of a resource by assuming a fixed number of copies of that res
ource. However\, in several situations\, it is not the case that the numbe
r of copies of a resource are fixed. Rather\, additional copies of a resou
rce might be procured but at a possibly high cost. The limitation of the r
esource is therefore\, due to the cost associated in procuring additional
copies. Yet\, additional copies can be procured.\n\nHighlight of the resul
ts:\n\nIn a work with Avrim Blum\, Anupam Gupta and Yishay Mansour\, we in
itiate the study of resource allocation in the more realistic modeling of
limitation where each resource has an increasing procurement cost. The res
ources need to be allocated amongst agents to maximizes welfare. In this t
alk\, we describe a simple allocation scheme that achieves a constant fact
or approximation to the optimal social welfare for polynomial and logarith
mic procurement cost curves. Furthermore\, for arbitrary procurement cost
curves\, we describe a scheme that achieves a logarithmic approximation to
optimal welfare.\n\nThe results are part of a work that appeared at Found
ations of Computer Science (FOCS)\, 2011 (joint work with Avrim Blum\, Anu
pam Gupta and Yishay Mansour).\n\nhttps://indico.tifr.res.in/indico/confer
enceDisplay.py?confId=1809
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1809
END:VEVENT
BEGIN:VEVENT
SUMMARY:Refined Approaches to Randomized Rounding
DTSTART;VALUE=DATE-TIME:20120103T060000Z
DTEND;VALUE=DATE-TIME:20120103T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1887@cern.ch
DESCRIPTION:Randomized rounding is a classic method to produce an integral
0/1 solution from a fractional one by interpreting the fractions as proba
bilities. However\, in many situations this rounding is too naive and lose
s various nice properties that the fractional solution may have possessed.
We will survey various dependent rounding approaches developed in recent
years that achieve the benefits of randomized rounding while maintaining o
ther desirable properties. In particular\, we will see how several recent
results such as sub-logarithmic approximation for asymmetric TSP\, constru
ctive algorithms for discrepancy minimization and additive guarantees for
degree bounded spanning trees can be viewed from this lens.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=1887
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1887
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fighting Boolean Circuits: New Correlation Bounds and Pseudorandom
Generators
DTSTART;VALUE=DATE-TIME:20120104T060000Z
DTEND;VALUE=DATE-TIME:20120104T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1906@cern.ch
DESCRIPTION:The general problem of proving limitations on what efficient a
lgorithms can accomplish has been the subject of much research over the pa
st four decades. In this talk\, we consider the specific question of bound
ing the power of Boolean circuits. We discuss three generic questions: tho
se of proving lower bounds\, correlation bounds\, and constructing Pseudor
andom generators (PRGs) for different classes of Boolean circuits. We look
at the major results in these directions and the important questions that
need to be tackled next.\n\nIn the process\, we motivate the question of
proving correlation bounds and constructing PRGs for bounded-depth circuit
s. We then describe our own results: correlation bounds for AC^0 with a fe
w symmetric gates\, and explicit PRGs against read-once ACC^0 (this is joi
nt work with Shachar Lovett (IAS\, Princeton) and Dmitry Gavinsky (NEC Res
earch Labs\, Princeton)).\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=1906
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1906
END:VEVENT
BEGIN:VEVENT
SUMMARY:Short Proofs of the Quantum Substate Theorem
DTSTART;VALUE=DATE-TIME:20120109T060000Z
DTEND;VALUE=DATE-TIME:20120109T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1924@cern.ch
DESCRIPTION:The Quantum Substate Theorem due to Jain\, Radhakrishnan\, and
(2002) gives us a powerful operational interpretation of relative entropy
\, in fact\, of the observational divergence of two quantum states\, a qua
ntity that is related to their relative entropy. Informally\, the theorem
states that if the observational divergence between two quantum states rho
\, sigma is small\, then there is a quantum state rho' close to rho in tra
ce distance\, such that rho' when scaled down by a small factor becomes a
substate of sigma. We present new proofs of this theorem. The resulting st
atement is optimal up to a constant factor in its dependence on observatio
nal divergence. In addition\, the proofs are both conceptually simpler and
significantly shorter than the earlier proof (this is joint work with Rah
ul Jain\, National University of Singapore).\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=1924
LOCATION:Colaba Campus A-212 (STCS Semin Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1924
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mitochondrial Variability
DTSTART;VALUE=DATE-TIME:20120110T060000Z
DTEND;VALUE=DATE-TIME:20120110T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1872@cern.ch
DESCRIPTION:In this talk I'll discuss our work on mitochondrial variabilit
y. Mitochondria serve as power stations for the cell and a conventional vi
ew has treated them as isolated organelles. In fact they are very dynamic:
networking\, fissioning\, fusing\, being consumed and being generated. We
recently found a surprising experimental connection between a marked vari
ability in the average rate of transcription of cells in a population and
a marked variability in their mitochondrial content. I'll discuss our math
ematical and computational attempts to summarize this data. They suggest t
he existence of a partly-heritable factor which is connected to cell-to-ce
ll variability. I'll then move on to discussing how the genetic content of
mitochondria might vary through time and make connections with disease.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1872
LOCATION:Colaba Campus A-269 (DAA Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1872
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Optimal Lower Bound for File Maintenance
DTSTART;VALUE=DATE-TIME:20120113T053000Z
DTEND;VALUE=DATE-TIME:20120113T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1931@cern.ch
DESCRIPTION:In the file maintenance problem\, $n$ integer items from the s
et ${1\,ldots\,r}$ are to be stored in an array of size $mgeq n$. The item
s are presented sequentially in an arbitrary order and must be stored in t
he array in sorted order (but not necessarily in consecutive locations in
the array). Each new item is stored before the next arrives. If $rleq m$ t
hen we can simply store item $j$ in location $j$\, but if $r>m$ then we ma
y have to shift the location of stored items to make space for a newly arr
ived item. The algorithm is charged each time an item is stored or moved t
o a new location. The goal is to minimize the total number of such moves.
The problem is non-trivial for $n leq mleq r$.\n\nIn the case that $m=Cn$
for some constant $C>1$\, there is an algorithm due to Dan Willard (buildi
ng on work of Itai\, Konheim and Rodeh) that stores each item at maximum c
ost at most $O((log n)^2)$ per item. In this paper we show that this is op
timal (even in the amortized sense): for any algorithm there is a sequence
of $n$ items that may require cost $Omega(n (log n)^2)$ (this is joint wo
rk with Jan Bulanek and Michal Koucky).\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=1931
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1931
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Church Synthesis Problem with Metric
DTSTART;VALUE=DATE-TIME:20120123T090000Z
DTEND;VALUE=DATE-TIME:20120123T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1975@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
1975
LOCATION:Colaba Campus
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1975
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Church Synthesis Problem with Metric
DTSTART;VALUE=DATE-TIME:20120123T090000Z
DTEND;VALUE=DATE-TIME:20120123T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1976@cern.ch
DESCRIPTION:Church's Problem asks for the construction of a procedure whic
h\, given a logical specification f(I\,O) between input strings I and outp
ut strings O\, determines whether there exists an operator F that implemen
ts the specification in the sense that f(I\,F(I)) holds for all inputs I.
Buchi and Landweber gave a procedure to solve Church's problem for MSO spe
cifications and operators computable by finite-state automata.\n\nWe consi
der extensions of Church's problem in two orthogonal directions: (i) we ad
dress the problem in a more general logical setting\, where not only the s
pecifications but also the solutions are presented in a logical system\; (
ii) we consider not only the canonical discrete time domain of the natural
numbers\, but also the continuous domain of reals.\n\nWe show that for ev
ery fixed bounded length interval of the reals\, Church's problem is decid
able when specifications and implementations are described in the monadic
second-order logics over the reals with order and the +1 function (joint w
ork with Mark Jenkins\, Joel Ouaknine and James Worrell).\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=1976
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1976
END:VEVENT
BEGIN:VEVENT
SUMMARY:Engineering Exquisite Nanoscale Behavior with DNA
DTSTART;VALUE=DATE-TIME:20120124T090000Z
DTEND;VALUE=DATE-TIME:20120124T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1965@cern.ch
DESCRIPTION:Self-assembly is a pervasive natural phenomenon that gives ris
e to complex structures and functions. It describes processes in which a d
isordered system of components form organized structures as a consequence
of specific\, local interactions among the components themselves\, without
any external direction. Biological self-assembled systems\, evolved over
billions of years\, are more intricate\, more energy efficient and more fu
nctional than anything researchers have currently achieved at the nanoscal
e. A challenge for human designed physical self-assembled systems is to ca
tch up with mother nature. In this talk\, we argue\, through examples of o
ur contributions in the field\, that DNA is an apt material to meet this c
hallenge. I will present: (1) 3D self-assembled nanostructures\, (2) Illus
trations of the simplicity and power of toehold-mediated strand displaceme
nt interactions and (3) Algorithmic constructs in the tile assembly model.
\n\nBio: \n\nNikhil Gopalkrishnan is a PhD candidate and James B. Duke Fel
low at the Duke Computer Science department working with Prof. John Reif.
His research interests are in self-assembly based models of computation\,
computing with DNA molecules and experimental DNA self-assembly. Nikhil ha
s published papers in ICALP\, SICOMP\, Algorithmica\, Journal of the Royal
Society Interface and DNA Computing & Molecular Programming. He has also
co-authored chapters in three books reviewing research in his field.\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1965
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1965
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximate Identities
DTSTART;VALUE=DATE-TIME:20120124T113000Z
DTEND;VALUE=DATE-TIME:20120124T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1980@cern.ch
DESCRIPTION:I will explain the role played by infinitely differentiable fu
nctions with compact support on the real line. Suitably scaled such functi
ons are very close (with respect to a suitable norm) to being identity ele
ments for the binary operation of convolution of functions. If time permit
s\, as an application I will sketch the proof of the famous Weirstrass app
roximation theorem in real analysis.\n\nOnly knowledge of elementary calcu
lus will be assumed.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=1980
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1980
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum Query Complexity of State Conversion
DTSTART;VALUE=DATE-TIME:20120125T103000Z
DTEND;VALUE=DATE-TIME:20120125T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1967@cern.ch
DESCRIPTION:State conversion generalizes query complexity to the problem o
f converting between two input-dependent quantum states by making queries
to the input. We characterize the complexity of this problem by introducin
g a natural information-theoretic norm that extends the Schur product oper
ator norm. The complexity of converting between two systems of states is g
iven by the distance between them\, as measured by this norm. In the speci
al case of function evaluation\, the norm is closely related to the genera
l adversary bound\, a semi-definite program that lower-bounds the number o
f input queries needed by a quantum algorithm to evaluate a function. We t
hus obtain that the general adversary bound characterizes the quantum quer
y complexity of any function whatsoever. This generalizes and simplifies t
he proof of the same result in the case of boolean input and output. Also
in the case of function evaluation\, we show that our norm satisfies a rem
arkable composition property\, implying that the quantum query complexity
of the composition of two functions is at most the product of the query co
mplexities of the functions\, up to a constant. In this talk we will focus
more on designing new quantum query algorithms using classical tools.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1967
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1967
END:VEVENT
BEGIN:VEVENT
SUMMARY:Wireless Scheduling with Partial Information - Low Delay and High
Throughput Scheduling Algorithms
DTSTART;VALUE=DATE-TIME:20120130T053000Z
DTEND;VALUE=DATE-TIME:20120130T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1978@cern.ch
DESCRIPTION:Modern wireless networks support increasing amounts of data tr
affic over a wide range of services\, e.g. high-speed file transfer\, real
time audio/video and peer-to-peer applications. Effective scheduling is r
equired in wireless systems to deliver high performance\, with many users
and limited system resources. Together with maximizing user throughput\, s
cheduling must keep packet delays down to a minimum in order to reliably s
upport delay-sensitive applications like audio/video streaming and gaming.
The problem is especially compounded when\, due to limited feedback resou
rces in the wireless system\, only partial network state information is av
ailable to the scheduler. In this talk\, we consider wireless scheduling w
here channel state information can be obtained from only subsets of users.
For this setting\, we present joint subset-selection and user-scheduling
algorithms that maximize throughput and minimize the likelihood of long qu
eues. In particular\, these scheduling algorithms provably achieve the bes
t exponential decay rate (large-deviations rate function)\, among all sche
duling algorithms using partial channel state\, of the tail of the longest
queue in the system. Time permitting\, I will outline other recent work a
nd new directions in scheduling with information delays/mismatches\, the s
peed of controlled network information dissemination\, and learning in dyn
amic environments.\n\nBio: Aditya Gopalan received the B.Tech. and M.Tech.
degrees in Electrical Engineering from the Indian Institute of Technology
(IIT) Madras\, Chennai\, in 2006\, and the PhD in Electrical Engineering
from The University of Texas at Austin in 2011. He was a summer intern at
the Corporate Research and Development Center\, Qualcomm Inc.\, in 2009. H
e is currently a Post Doctoral Research Fellow in the Faculty of Electrica
l Engineering at the Technion- Israel Institute of Technology\, Haifa\, Is
rael. His current research interests include network algorithms\, stochast
ic control\, and machine learning.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=1978
LOCATION:Colaba Campus A-269 (DAA Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1978
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fault Tolerance in Distributed Systems: A Coding-Theoretic Approac
h
DTSTART;VALUE=DATE-TIME:20120206T103000Z
DTEND;VALUE=DATE-TIME:20120206T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1977@cern.ch
DESCRIPTION:Distributed systems are rapidly increasing in importance due t
o the need for scalable computation on huge volumes of data. This fact is
reflected in many distributed applications such as Amazon’s cloud comput
ing service\, Google’s BigTable or Apache’s Hadoop framework. Replicat
ion is the prevalent solution for fault tolerance in these systems. This s
olution\, though simple\, is wasteful in terms of both the space and infra
structure costs required to host the backups. We present a new paradigm to
solve this problem\, broadly referred to as fusion\, that combines the op
erational efficiency of replication with the space efficiency of coding th
eory. To make our techniques generally applicable\, we describe fusion in
two separate contexts: finite state machines and infinite state machines.\
n\nFor finite state machines\, we present a polynomial time algorithm to g
enerate efficient backup machines. For infinite state machines\, we consid
er programs that host large data structures such as linked lists\, stacks\
, vectors and maps. Using a combination of erasure codes and selective rep
lication\, we present an algorithm to generate efficient backup data struc
tures. We prove the minimality of our schemes and also provide experimenta
l results that confirm the same. Finally\, we present a fusion-based desig
n for fault tolerance in two real world applications: Amazon’s key-value
store\, Dynamo\, and Google's Map Reduce framework\, that is much more ef
ficient than the current replication-based approach.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=1977
LOCATION:Colaba Campus AG-66 (Lecture Threatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1977
END:VEVENT
BEGIN:VEVENT
SUMMARY:How to Sort a Train
DTSTART;VALUE=DATE-TIME:20120210T060000Z
DTEND;VALUE=DATE-TIME:20120210T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-1987@cern.ch
DESCRIPTION:This talk attempts to provide an introductory survey of method
s for rearranging cargo trains in railway yards from an algorithmic perspe
ctive\, and to highlight open problems in this area. The reported work has
been done over the years with Holger Flier\, Riko Jacob\, Jens Maue\, Mat
us Mihalak\, and Marc Nunkesser.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=1987
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=1987
END:VEVENT
BEGIN:VEVENT
SUMMARY:Large-scale Situation Awareness with Camera Networks and Multimoda
l Sensing
DTSTART;VALUE=DATE-TIME:20120220T060000Z
DTEND;VALUE=DATE-TIME:20120220T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2009@cern.ch
DESCRIPTION:Sensors of various modalities and capabilities\, especially ca
meras\, have become ubiquitous in our environment. Their intended use is w
ide ranging and encompasses surveillance\, transportation\, entertainment\
, education\, healthcare\, emergency response\, disaster recovery\, and th
e like. Technological advances and the low cost of such sensors enable dep
loyment of large-scale camera networks in large metropolis such as London
and New York. Multimedia algorithms for analyzing and drawing inferences f
rom video and audio have also matured tremendously in recent times. Despit
e all these advances\, large-scale reliable systems for media-rich sensor-
based applications\, often classified as situation awareness applications\
, are yet to become commonplace. Why is that? There are several forces at
work here. First of all\, the system abstractions are just not at the righ
t level for quickly prototyping such applications in the large. Second\, w
hile Moore’s law has held true for predicting the growth of processing p
ower\, the volume of data that applications are called upon to handle is g
rowing similarly\, if not faster. Enormous amount of sensing data is conti
nually generated for real-time analysis in such applications. Further\, du
e to the very nature of the application domain\, there are dynamic and dem
anding resource requirements for such analyses.\n\nThe lack of right set o
f abstractions for programming such applications coupled with their data i
ntensive nature have hitherto made realizing reliable large-scale situatio
n awareness applications difficult. Incidentally\, situation awareness is
a very popular but ill-defined research area that has attracted researcher
s from many different fields. In this talk\, I will adopt a systems perspe
ctive and consider the components that are essential in realizing a fully
functional situation awareness system.\n\nBio: Kishore Ramachandran receiv
ed his Ph. D. in Computer Science from the University of Wisconsin\, Madis
on in 1986\, and has been on the faculty of Georgia Tech since then. Curre
ntly\, he is the Director of Samsung Tech Advanced Research (STAR) center\
, Director of Korea Programs\, and Professor in the College of Computing a
t the Georgia Institute of Technology. For two years (July 2003 to August
2005) he served as the Chair of the Core Computing Division within the Col
lege of Computing. His fields of interest include parallel and distributed
systems\, computer architecture\, and operating systems. He has authored
over 100 technical papers and is best known for his work in Distributed Sh
ared Memory (DSM) in the context of the Clouds operating system\; and more
recently for his work in stream-based distributed programming in the cont
ext of the Stampede system. Currently\, he is leading two inter-related pr
ojects\, ASAP/TC that deals with camera sensor networks for situation awar
eness and Web on Demand that deals with system software stack for enabling
transient social networking on mobile platforms. He has so far graduated
22 Ph.D. students who are well placed in academia and industries. He is cu
rrently advising 8 Ph.D. students. He is the recipient of an NSF PYI Award
in 1990\, the Georgia Tech doctoral thesis advisor award in 1993\, the Co
llege of Computing Outstanding Senior Research Faculty award in 1996\, the
College of Computing Dean's Award in 2003\, the College of Computing Will
iam ``Gus'' Baird Teaching Award in 2004\, and the 2009 “Freeman Award
” for entrepreneurship in the College of Computing.\n\nhttps://indico.ti
fr.res.in/indico/conferenceDisplay.py?confId=2009
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2009
END:VEVENT
BEGIN:VEVENT
SUMMARY:Spectrum Pricing Games in Cognitive Radio Networks
DTSTART;VALUE=DATE-TIME:20120227T053000Z
DTEND;VALUE=DATE-TIME:20120227T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2021@cern.ch
DESCRIPTION:We study price competition among primaries in a Cognitive Radi
o Network (CRN) with multiple primaries and secondaries located in a large
region. In every slot\, each primary has unused bandwidth with some proba
bility\, which may be different for different primaries. Also\, there may
be a random number of secondaries. A primary can lease out its unused band
width to a secondary in exchange for a fee. Each primary tries to attract
secondaries by setting a lower price for its bandwidth than the other prim
aries. Radio spectrum has the distinctive feature that transmissions at ne
ighboring locations on the same channel interfere with each other\, wherea
s the same channel can be used at far-off locations without mutual interfe
rence. So in the above price competition scenario\, each primary must join
tly select a set of mutually non-interfering locations within the region (
which corresponds to an independent set in the conflict graph representing
the region) at which to offer bandwidth and the price at each location. W
e analyze this price competition scenario as a game and seek a Nash Equili
brium (NE). For the game at a single location\, we explicitly compute a NE
and prove its uniqueness. Also\, for the game at multiple locations\, we
identify a class of conflict graphs\, which we refer to as mean valid grap
hs\, such that the conflict graphs of a large number of topologies that co
mmonly arise in practice are mean valid. We explicitly compute a NE in mea
n valid graphs and show that it is unique in the class of NE with symmetri
c independent set selection strategies of the primaries.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=2021
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2021
END:VEVENT
BEGIN:VEVENT
SUMMARY:One Shot Slepian-Wolf
DTSTART;VALUE=DATE-TIME:20120228T090000Z
DTEND;VALUE=DATE-TIME:20120228T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2035@cern.ch
DESCRIPTION:Slepian and Wolf gave the rate-region for the distributed enco
ding of correlated and memoryless sources when the number of copies of sou
rce output is unlimited. We give one-shot rate region for the Slepian-Wolf
protocol when a single copy of the source output is available. Our result
s are asymptotically optimal\, i.e.\, they yield the same rate region as t
he Slepian-Wolf in the limit of unlimited copies. We also give an upper bo
und for the one-shot encoding of a single source that is different from th
e one given by Renner and Wolf.\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=2035
LOCATION:Colaba Campus
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2035
END:VEVENT
BEGIN:VEVENT
SUMMARY:Two Characterizations of General Ergodic Markov Chain Families wit
h Small First Passage Time
DTSTART;VALUE=DATE-TIME:20120229T083000Z
DTEND;VALUE=DATE-TIME:20120229T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2030@cern.ch
DESCRIPTION:Markov chain is a stochastic process which has been found effe
ctive in modelling a wide variety of situations and phenomena in computer
science. In many such applications\, the first passage times of certain st
ates of Markov chains become matters of interest. One such example is rand
omized search heuristics for solving combinatorial optimization problems.
The execution of many such heuristics can be modelled by a Markov chain wh
ere each state corresponds to a feasible solution\, and the target state i
s the one corresponding to minimum cost (for minimization problem). So the
run time of such an algorithm is the first passage time of the target sta
te. Other examples include gamble and situations that can be modelled as g
ambles\, randomized routing algorithms in computer networks\, etc. The the
sis focusses on characterizing families of ergodic Markov chains having sm
all first passage time of some goal state. To appropriately define `small'
\, we consider Markov chains parametrized by binary strings and define `sh
ort first passage time' to be at most a fixed polynomial in the size of th
e string that gives rise to the chain. We introduce two notions of 'succes
s' as we call it\, and show them to be equivalent. The thesis presents two
characterizations of success. The first one is that the ratio of the ergo
dic flow out of each set of states not containing the goal state to the ca
pacity of the state is at least some inverse polynomial in the size of the
string giving rise to the chain. The second characterization\, which is e
asily seen to be a sufficient condition\, is that the family of Markov cha
ins is rapidly mixing and the stationary probability of the goal state is
at least some inverse polynomial in the size of the generating string. Thu
s rapid mixing and first passage time turn out to be closely related. Fina
lly we illustrate the applicability of our results by giving alternative p
roofs of some known results due to Wegener on performance of the Metropoli
s algorithm on the minimum spanning tree problem for connecting triangles.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2030
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2030
END:VEVENT
BEGIN:VEVENT
SUMMARY:How to Achieve Consensus in Society?
DTSTART;VALUE=DATE-TIME:20120302T093000Z
DTEND;VALUE=DATE-TIME:20120302T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2060@cern.ch
DESCRIPTION:I will introduce a very simple social interaction model of inf
ormation transmission\, opinion formation and consensus formation by De Gr
oot. We will study sufficient and necessary conditions to converge to a co
nsensus in very simple networks. Building on this\, I will introduce how w
e can modify the model to make it resemble real life networks. If time per
mits\, I will try to show how the convergence is affected when interaction
is noisy.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=2060
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2060
END:VEVENT
BEGIN:VEVENT
SUMMARY:How Much Training is Needed in Reciprocal Multiple Antenna Wireles
s Links?
DTSTART;VALUE=DATE-TIME:20120305T060000Z
DTEND;VALUE=DATE-TIME:20120305T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2020@cern.ch
DESCRIPTION:In this talk\, we discuss the problem of data transmission in
a single input\, multiple output wireless communication system with channe
l estimation at both transmitter and receiver\, from a Diversity- Multiple
xing gain Trade-off (DMT) perspective. We consider a two-way communication
system with time-division duplexing\, which makes it possible to acquire
Channel State Information (CSI) at the transmitter through reverse channel
training\, i.e.\, by sending a training signal from the receiver to the t
ransmitter. We propose several schemes for reverse channel training where
the training signal is dependent on the receiver's knowledge of CSI (which
could be acquired\, for example\, through an initial forward-link trainin
g phase). We explicitly account for the time-overhead in the training sche
mes\, and show that our proposed reverse training and power control scheme
s offer a significant improvement in terms of DMT compared to conventional
channel-agnostic training schemes. Thus\, we argue that our proposed reve
rse channel training schemes are a promising technique to improve the reli
ability and throughput of current and future wireless communication system
s.\n\nBio: Dr. Chandra R. Murthy is an assistant professor at IISc's ECE d
epartment. His interests are in digital signal processing\, information th
eory\, estimation theory\, and their applications in the optimization of M
IMO\, OFDM and CDMA wireless communication systems.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=2020
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2020
END:VEVENT
BEGIN:VEVENT
SUMMARY:Bourgain's Theorem
DTSTART;VALUE=DATE-TIME:20120309T093000Z
DTEND;VALUE=DATE-TIME:20120309T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2066@cern.ch
DESCRIPTION:Problems in Metric embedding involve mapping a set (for our pu
rposes\, finite) of points from one metric space to another\, while preser
ving pairwise distances. Of late\, this has attracted much research intere
st in computer science as a tool for designing approximation algorithms. W
e shall see the proof of a fundamental theorem due to Bourgain\, which sta
tes that n points in *any* metric space on can be mapped into l_1 space (R
^d with the l_1 norm between points)\, while increasing pairwise distances
by a factor of at most O(log(n)).\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=2066
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2066
END:VEVENT
BEGIN:VEVENT
SUMMARY:Simple Channel Coding Bounds
DTSTART;VALUE=DATE-TIME:20120316T093000Z
DTEND;VALUE=DATE-TIME:20120316T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2072@cern.ch
DESCRIPTION:New channel coding converse and achievability bounds are deriv
ed for a single use of an arbitrary channel. Both bounds are expressed usi
ng a quantity called the “smooth 0-divergence”\, which is a generaliza
tion of R´enyi’s divergence of order 0.\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=2072
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2072
END:VEVENT
BEGIN:VEVENT
SUMMARY:Is Machine Learning Easy?
DTSTART;VALUE=DATE-TIME:20120320T090000Z
DTEND;VALUE=DATE-TIME:20120320T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2073@cern.ch
DESCRIPTION:Machine learning is a flourishing field with a plethora of alg
orithmic techniques. In almost all cases\, we have no provable guarantees
on the performance of these algorithms/heuristics ---neither on their runn
ing time\, nor on the quality of solutions they return. In fact in many se
ttings\, the underlying algorithmic problems are provably intractable (e.g
.\, NP-hard or worse)! This talk will argue that this seeming intractabili
ty may arise because many models used in machine learning are more general
than they need to be.\n\nThis point will be illustrated using our recent
results on designing provable algorithms for nonnegative matrix factorizat
ion (NMF)\, a well-known NP-hard problem that is popular in many areas of
machine learning. We present the first algorithms that can find factorizat
ions of rank r in polynomial time (in the matrix size\, not r). We also in
troduce a natural new subcase of NMF called "separable NMF" which we think
is the proper notion to use in many contexts. We give fully polynomial-ti
me algorithms for separable NMF. We go on to use these algorithms for more
general problems involving learning of topic models (such as the wellknow
n Latent Dirichlet Allocation of Blei et al).\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=2073
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2073
END:VEVENT
BEGIN:VEVENT
SUMMARY:Weak Epsilon Nets
DTSTART;VALUE=DATE-TIME:20120326T060000Z
DTEND;VALUE=DATE-TIME:20120326T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2105@cern.ch
DESCRIPTION:Given a set $P$ of $n$ points in $R^d$\, a weak $epsilon$-net
of $P$ with respect to convex sets in a subset of $R^d$ that intersects ev
ery convex set containing an $epsilon$-fraction of the points in $P$. Find
ing optimal bounds for the size of a weak $epsilon$-net is a fundamental p
roblem in discrete and convex geometry and has been an open problem for 25
years. While we have a very good understanding of strong $epsilon$-nets\,
there is huge gap in the known bounds for weak $epsilon$-nets. The curren
t upper bound is $O(epsilon^{-d} polylog epsilon^{-1})$ while the lower bo
und is $Omega( epsilon^{-1} log^{d-1} epsilon^{-1})$. Both weak and strong
$epsilon$-nets have been used as rounding procedures for various set-cove
r type problems. Better bounds for their sizes therefore translate to impr
oved bounds on integrality gaps of these problems.\n\nI will describe the
known results about the weak $epsilon$-net problem and then discuss some p
artial results and approaches towards improving the bounds.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=2105
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2105
END:VEVENT
BEGIN:VEVENT
SUMMARY:Epsilon-nets and its Relatives
DTSTART;VALUE=DATE-TIME:20120327T103000Z
DTEND;VALUE=DATE-TIME:20120327T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2069@cern.ch
DESCRIPTION:Epsilon nets and approximations are among the most fundamental
notions of sampling. For a given set system and a parameter $epsilon$\, a
n $epsilon$-net is a transveral for the sets of size at least $epsilon n$\
, $n$ being the size of the ground set. The idea originated in the context
of learning theory and was introduced by Haussler and Welzl in 1987. With
a delicate use of random sampling\, they showed that any set system of fi
nite VC dimension has an $epsilon$-net of small size independent of the si
ze of the set system - and moreover a random sample of this size is an $ep
silon$-net with high probability. Epsilon nets are now an important tool t
hat has found applications in many areas including computational geometry\
, approximation algorithms\, discrepancy theory and statistics. Very simpl
e and deterministic constructions were found by Matousek in 1991. Determin
istic algorithms for computing $epsilon$-nets have turned them into a gene
ral tool for divide & conquer and for derandomization - many of the sophis
ticated data structures and algorithms are based on $epsilon$-nets. They p
ower tools like cuttings and simplicial partitions that are among the fine
st algorithmic tools in geometry. Besides algorithmic applications\, they
have have been used for the proof of other combinatorial results. Recent r
esearch on Epsilon nets and related problems has revealed further connecti
ons to other areas of mathematics.\n\nIn this talk I will give an introduc
tion to $epsilon$-nets and the various related problems that I have worked
on. I will describe my results on the construction of $epsilon$-nets\, re
lated geometric set cover problems\, enumeration problems and weak $epsilo
n$-nets.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
2069
LOCATION:Colaba Campus A-269 (DAA Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2069
END:VEVENT
BEGIN:VEVENT
SUMMARY:In Search of Boundaries
DTSTART;VALUE=DATE-TIME:20120330T093000Z
DTEND;VALUE=DATE-TIME:20120330T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2109@cern.ch
DESCRIPTION:Suppose we have a number of items\, each item belongs to eithe
r class 1 or class 2. Given the items only\, we want to build an algorithm
that will automatically classify these items into the respective classes
after it has learnt from some examples. This is a simple classification pr
oblem. One of the earliest algorithms for classification is the Rosenblatt
's Perceptron. The Perceptron algorithm tries to classify the items by bui
lding a hyperplane between the items of the two classes. In this talk we w
ill prove a theorem due to Novikoff which asserts that\, under some condit
ions\, if the two classes can be separated by hyperplanes then the Percept
ron algorithm will find one such hyperplane.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=2109
LOCATION:Colaba Campus A-269 (DAA Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2109
END:VEVENT
BEGIN:VEVENT
SUMMARY:Automated Verification of Cryptographic Protocols
DTSTART;VALUE=DATE-TIME:20120413T043000Z
DTEND;VALUE=DATE-TIME:20120413T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2116@cern.ch
DESCRIPTION:The widespread use of internet has raised serious concerns of
privacy and trust. In order to address these concerns\, cryptographic prot
ocols are widely used. A cryptographic protocol is a distributed program t
hat uses cryptographic primitives to ensure security over an untrusted net
work. However\, the design of cryptographic protocols has proven to be err
or-prone and several errors have been found. Thus\, there is a need for bu
ilding scalable tools for automatically verifying security of cryptographi
c protocols. The complexity of cryptographic protocols as well as the desi
red security guarantees presents unique challenges to verification of cryp
tographic protocols. We illustrate these challenges within the context of
verifying game-based and equivalence-based properties of cryptographic pro
perties.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
2116
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2116
END:VEVENT
BEGIN:VEVENT
SUMMARY:An FPTAS for Optimizing a Class of Low-rank Functions Over a Polyt
ope
DTSTART;VALUE=DATE-TIME:20120416T103000Z
DTEND;VALUE=DATE-TIME:20120416T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2074@cern.ch
DESCRIPTION:We present a fully polynomial time approximation scheme (FPTAS
) for optimizing a very general class of non-linear functions of low rank
over a polytope. Our approximation scheme relies on constructing an approx
imate Pareto-optimal front of the linear functions which constitute the gi
ven low-rank function. In contrast to existing results in the literature\,
our approximation scheme does not require the assumption of quasi-concavi
ty on the objective function. Our technique can also be used to obtain an
FPTAS for combinatorial optimization problems with non-linear objective fu
nctions\, for example when the objective is a product of a fixed number of
linear functions. We also show that it is not possible to approximate the
minimum of a general concave function over the unit hypercube to within a
ny factor\, unless P = NP. We prove this by showing a similar hardness of
approximation result for supermodular function minimization\, a result tha
t may be of independent interest (this is joint work with Andreas S. Schul
z).\n\n[Available from: http://dx.doi.org/10.1007/s10107-011-0511-x ]\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2074
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2074
END:VEVENT
BEGIN:VEVENT
SUMMARY:Logics and Automata for Classical and Real Time Unambiguous Langua
ges
DTSTART;VALUE=DATE-TIME:20120417T103000Z
DTEND;VALUE=DATE-TIME:20120417T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2138@cern.ch
DESCRIPTION:Regular languages exhibit considerable structure. The characte
rization of various classes of regular languages using classical and tempo
ral logics\, automata and varieties of monoids\, as an attempt towards uni
fication of diverse frameworks\, has been an important theme and area of r
esearch for many decades. In this thesis\, we investigate logic-automata c
onnections for unambiguous languages over finite words and also timed word
s.\n\nUnambiguous star-free regular languages (UL) originally introduced b
y Schutzenberger\, have various characterizations within classical logics
(FO2[\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=213
8
LOCATION:Colaba Campus A-414 (A-Block Conference Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2138
END:VEVENT
BEGIN:VEVENT
SUMMARY:Functional Encryption Systems From Hard Lattice Problems
DTSTART;VALUE=DATE-TIME:20120420T050000Z
DTEND;VALUE=DATE-TIME:20120420T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2078@cern.ch
DESCRIPTION:Data security challenges faced in the modern world demand func
tionality from encryption systems that traditional public key cryptography
falls far short in delivering.\n\nTake the example of cloud computing\, a
paradigm which allows users to outsource their data storage and computing
needs to a powerful third party server such as Amazon. Though such a serv
ice is very useful\, users may be reluctant to trust third party servers w
ith sensitive data. Organizations utilizing these services must also ensur
e that their clients are secure from each other. At the same time\, meanin
gful functionality must be provided. For example a server storing medical
data might be required to grant users access to certain useful functions c
omputed on the entire user database\, such as the success rate of some med
ication for a given disease\, while making sure that individual medical pr
ivacy is not compromised.\n\nTo address these emerging needs\, a new parad
igm of encryption was recently put forward – Functional Encryption. In f
unctional encryption\, a user's secret key can be associated with its hold
er's credentials\, while the ciphertext can be associated with an access p
olicy. We may ask that decryption succeed if and only if the credentials s
atisfy the access policy.\n\nI will describe several special cases of func
tional encryption that we have constructed -- systems for the identity fun
ction (identity based encryption or IBE)\, threshold function (fuzzy IBE)
and linear functions. I will describe ongoing work to provide a general fr
amework for these constructions and challenges faced in supporting more ge
neral functions. The technical tool we use in these constructions is the w
orst-case hardness of lattice problems. Lattices have traditionally been u
sed in cryptography for breaking cryptosystems and their use in building c
ryptosystems is surprising and elegant.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=2078
LOCATION:Colaba Campus D-405 (D-Block Conference Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2078
END:VEVENT
BEGIN:VEVENT
SUMMARY:Geometric Searching
DTSTART;VALUE=DATE-TIME:20120420T103000Z
DTEND;VALUE=DATE-TIME:20120420T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2148@cern.ch
DESCRIPTION:To describe searching in its simplest abstract setting\, suppo
se we have some accumulated data (called the 'file') and some new data ite
m (called the 'sample'). Then\, searching consists of relating the sample
to the file. As Knuth pointed out\, searching simply means locating the ap
propriate record (or records) in a given collection of records.\n\nAlthou
gh all searching problems share some basic common feature\, the geometric
setting gives rise to questions of a somewhat unique flavour. First of all
\, in geometric applications\, files are primarily not just 'collections'\
, as in several other information processing activities. Rather\, they are
frequently representations of more complex structures\, such as polygons\
, polyhedra etc. Even the case of a collection of segments in the plane ap
pears deceptively unstructured\, since each segment is associated with its
end-point co-ordinates\, which are implicitly related by the metric struc
ture of the plane. Moreover\, the fact that co-ordinates of points are tho
ught of as real numbers entails that a search may not exactly find an item
in the file matching the sample\, but rather it will locate the latter in
relation to the file. This provides an additional point of contrast betwe
en geometric searching and conventional searching.\n\nNow\, suppose we hav
e a collection of geometric data\, and we want to know if it possesses a c
ertain property (say convexity). In the simplest case\, the query will be
asked only once\, in which case it would be wasteful to do any preconditio
ning in the hope of speeding up future queries. A one-time query of this t
ype will be referred to as 'single-shot'. Many times\, however\, queries w
ill be performed repeatedly on the same file. Such queries will be referre
d to as 'repetitive-mode' queries. In this case\, it may be worthwhile to
arrange the information into an organized structure to facilitate searchin
g. This can be accomplished only at some additional expense though. Then\,
our analysis must focus on four separate cost measures\, namely -\n\n1) Q
uery Time\n2) Preprocessing Time\n3) Storage Space\n4) Update Time\n\nIn m
y talk\, I will try to illustrate the various trade-offs among these cost
measures through the following 2 problems -\n\nA) RANGE SEARCHING : Given
N points in the plane\, how many lie in a given rectangle with sides paral
lel to the co-ordinate axes? That is\, how many points (x\,y) satisfy a \n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2148
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2148
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Little Advice can be very Helpful
DTSTART;VALUE=DATE-TIME:20120423T060000Z
DTEND;VALUE=DATE-TIME:20120423T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2126@cern.ch
DESCRIPTION:Proving superpolylogarithmic lower bounds for dynamic data str
uctures has remained an open problem despite years of research. Recently\,
Patrascu proposed an exciting new approach for breaking this barrier via
a two player communication model in which one player gets private advice a
t the beginning of the protocol. He gave reductions from the problem of so
lving an asymmetric version of set-disjointness in his model to a diverse
collection of natural dynamic data structure problems in the cell probe mo
del. He also conjectured that\, for any hard problem in the standard two-p
arty communication model\, the asymmetric version of the problem is hard i
n his model\, provided not too much advice is given.\n\nWe prove several s
urprising results about his model. We show that there exist Boolean functi
ons requiring linear randomized communication complexity in the two-party
model\, for which the asymmetric versions in Patrascu's model have determi
nistic protocols with exponentially smaller complexity. For set-disjointne
ss\, which also requires linear randomized communication complexity in the
two-party model\, we give a deterministic protocol for the asymmetric ver
sion in his model with a quadratic improvement in complexity. These result
s demonstrate that Patrascu's conjecture\, as stated\, is false. In additi
on\, we show that the randomized and deterministic communication complexit
ies of problems in his model differ by no more than a logarithmic multipli
cative factor.\n\nWe also prove lower bounds in some restricted versions o
f this model for natural functions such as set-disjointness and inner prod
uct. All of our upper bounds conform to these restrictions (joint work wit
h Jeff Edmonds\, Faith Ellen and Toniann Pitassi).\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=2126
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2126
END:VEVENT
BEGIN:VEVENT
SUMMARY:Nearest Neighbor Method in Simulation based Approximate Dynamic Pr
ogramming for Option Pricing
DTSTART;VALUE=DATE-TIME:20120424T083000Z
DTEND;VALUE=DATE-TIME:20120424T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2145@cern.ch
DESCRIPTION:As is well known\, options and other financial derivatives acq
uire their value from the underlying assets such as stocks and bonds and a
re traded extensively in the financial markets. We consider the problem of
pricing an American option\, that is an option with an early exercise fea
ture. This problem can be formulated as an optimal stopping time problem
for stochastic processes. In high dimensions\, typically no closed form so
lution is available for American option prices and numerical methods rely
ing on solving associated partial differential equations are also not viab
le. Recently there has been considerable research in developing simulation
methods to solve this optimal stopping time problem via approximate dynam
ic programming. We overview commonly used pricing methods that use nested
simulation and regression based techniques. We propose a new pricing algor
ithm wherein nearest neighbor estimator is used to estimate the continuati
on value functions required for the American option pricing. We derive asy
mptotic mean square error (MSE) of the option price estimator and find the
optimal parameter for the nearest neighbor estimator that minimizes this
MSE. This asymptotic MSE decays to zero as the allocated computational eff
ort increases to infinity. We also discuss the impact of dimensionality on
this rate of convergence.\n\nhttps://indico.tifr.res.in/indico/conference
Display.py?confId=2145
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2145
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimal Change of Measure for Model Selection and Efficient Simula
tion of Rare Event Probabilities with Financial Applications
DTSTART;VALUE=DATE-TIME:20120425T103000Z
DTEND;VALUE=DATE-TIME:20120425T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2143@cern.ch
DESCRIPTION:Change of measure techniques are ubiquitous in many areas of a
pplied probability including in model selection in finance and in efficien
t simulation of rare events.\n\nBroadly\, this technique amounts to changi
ng the underlying probability measure (the prior) to a new one (the poster
ior)\, so that the new probability measure may satisfy certain desirable p
roperties for the problem at hand. For example the posterior measure may b
e satisfying some constraints while being close in a specified sense to gi
ven probability measure\, or it may be efficient to simulate from the post
erior probability distribution to ascertain probability of some rare event
by simulation-estimation. In this thesis we develop change of measure in
two broad areas: Model selection for financial applications and simulatio
n-estimation of rare event probabilities.\n\nThe first abstracted problem
corresponds to finding a probability measure that minimizes the relative e
ntropy (also called I-divergence) with respect to a known measure while it
satisfies certain moment constraints on functions of underlying assets. W
e show that under I-divergence\, the optimal solution may not exist when t
he underlying assets have fat tailed distributions\, popular in financial
practice. We note that this drawback may be corrected if `polynomial-dive
rgence' is used. This divergence can be seen to be equivalent to the well
known (relative) Tsallis or (relative) Renyi entropy. We discuss existence
and uniqueness issues related to this new optimization problem. We also i
dentify the optimal solution structure under I-divergence as well as polyn
omial-divergence when the associated constraints include those on marginal
distribution of functions of underlying assets. These results are applied
to a simple problem of model calibration to options prices as well as to
portfolio modeling in Markowitz framework\, where we note that a reasonabl
e view that a particular portfolio of assets has heavy tailed losses may l
ead to fatter and more reasonable tail distributions of all assets.\n\nThe
second problem is further classified as estimation of large deviation pro
babilities and estimation of probability that sum of few random variables
exceeds a large threshold\, the latter is applied to pricing deep out of t
he money financial options. We exploit the saddle-point based representati
ons that exist for these rare quantities\, which rely on inverting the cha
racteristic functions of the underlying random vectors. We note that these
representations reduce the rare event estimation problem to evaluating ce
rtain integrals\, which may via importance sampling be represented as expe
ctations. Further\, the integrands have enough structure so that it is eas
y to identify and approximate the zero-variance importance sampling distri
bution to estimate these integrals. We identify such importance sampling m
easures and prove that asymptotically they possess strong efficiency prope
rties.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=21
43
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2143
END:VEVENT
BEGIN:VEVENT
SUMMARY:ICT Initiatives in Ethiopia
DTSTART;VALUE=DATE-TIME:20120426T103000Z
DTEND;VALUE=DATE-TIME:20120426T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2144@cern.ch
DESCRIPTION:Ethiopia\, as one of the developing nation\, is in the process
of implementing a number of ICT projects to make the Government electroni
cally enabled. Dr. Mesfin\, Director of E-Government Directorate at the M
inistry of Communication and Information Technology (MCIT)\, will give an
overview of the Information Communication Technology (ICT) initiatives in
Ethiopia. He will be talking on issues like Ethiopia in ICT International
Measures\, the Major ICT Initiatives\, the E-Government Strategy and on s
ome of the Challenges & Opportunities faced in implementation.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=2144
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2144
END:VEVENT
BEGIN:VEVENT
SUMMARY:EF Games for Expressiveness in Time
DTSTART;VALUE=DATE-TIME:20120427T093000Z
DTEND;VALUE=DATE-TIME:20120427T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2178@cern.ch
DESCRIPTION:The two key timed logics studied in literature are TPTL and MT
L. Their several sublogics are obtained by placing restrictions such as bo
unded interval constraints\, non-punctual constraints\, etc.\, With such
a variety of logics being studied\, there is a need to systematically comp
are them from an expressiveness point-of-view.\n\nAbout 20 years back\, Al
ur and Henzinger conjectured that TPTL[U\,S] is strictly more expressive t
han MTL[U\,S]. Fifteen years later\, a result by Bouyer et al. partially r
esolved this question by showing that MTL[U] is less expressive than TPTL[
U]. We proved the original conjecture in its full generality using EF game
s for MTL.\n\nEhrenfeucht-Fra\\"{\\i}ss\\'e games have been used in the st
udy of expressiveness of various classical and temporal logics. We extend
these to time and define MTL EF games. Essentially\, the MTL-EF theorem al
lows us to establish the inability of an MTL fragment in distinguishing tw
o timed behaviors. This technique gives remarkably simpler proofs for many
existing results on expressiveness and it also answers some long-standing
questions such as the Alur-Henzinger conjecture.\n\nhttps://indico.tifr.r
es.in/indico/conferenceDisplay.py?confId=2178
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2178
END:VEVENT
BEGIN:VEVENT
SUMMARY:Unbiased Coin Tossing With a Biased Coin
DTSTART;VALUE=DATE-TIME:20120504T093000Z
DTEND;VALUE=DATE-TIME:20120504T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2193@cern.ch
DESCRIPTION:I will discuss some algorithms for simulating a flip of an unb
iased coin by flipping a coin of unknown bias. We will discuss efficiency
of several algorithms\, where the expected number of flips is our measure
of efficiency. We will also outline the proof of a conjecture by Hoeffding
and Simons that there is no optimal algorithm.\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=2193
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2193
END:VEVENT
BEGIN:VEVENT
SUMMARY:When is it Optimal to Invest?
DTSTART;VALUE=DATE-TIME:20120511T093000Z
DTEND;VALUE=DATE-TIME:20120511T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2213@cern.ch
DESCRIPTION:In this talk\, we will see how to use variational inequalities
to solve a particular optimal stopping problem under certain conditions o
n the underlying diffusion processes.\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=2213
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2213
END:VEVENT
BEGIN:VEVENT
SUMMARY:Phase Shifting Masks for Microlithography
DTSTART;VALUE=DATE-TIME:20120518T093000Z
DTEND;VALUE=DATE-TIME:20120518T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2228@cern.ch
DESCRIPTION:I will introduce the problem of mask design in a lithographic
process from a systems perspective and show how it can be reduced to a non
-linear optimization problem. I will also show the benefit of phase shift
mask over the binary mask with some examples. Lastly\, I will show how a
complex mask can be replaced by successive real valued masks (double expos
ure theorem).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=2228
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2228
END:VEVENT
BEGIN:VEVENT
SUMMARY:Link Delay Tomography Via Independent Samples of Path Delay Measur
ements
DTSTART;VALUE=DATE-TIME:20120519T060000Z
DTEND;VALUE=DATE-TIME:20120519T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2191@cern.ch
DESCRIPTION:Link delay tomography is a practically viable approach to obta
in statistical information about the delay across each link in a network u
sing only path level measurements. In this talk\, we present a novel metho
d that can be used to estimate the complete distribution of the link delay
s upto any desired accuracy. The major highlight of this method is that\,
unlike all previous works\, it requires as input only a sequence of indepe
ndent samples of the end-to-end path delay measurements. The idea is to ap
proximate each link delay distribution using a generalized hyperexponentia
l distribution\, whose exponential stage parameters are known in advance\,
and focus on estimating the unknown mixing weights. These weights are obt
ained by solving a set of polynomial systems based on the moment generatin
g function of the end-to-end delays. For unique identifiability\, it is on
ly required that the network be 1-identifiable\; a condition which is esse
ntially true for all tree based networks.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=2191
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2191
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Information Theoretic Framework for Sensor Data Fusion for Rob
otics Applications
DTSTART;VALUE=DATE-TIME:20120523T090000Z
DTEND;VALUE=DATE-TIME:20120523T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2179@cern.ch
DESCRIPTION:In this talk I will present an information theoretic framework
for signallevel multimodal sensor data fusion. In particular I will focus
on the fusion of 3D lidar and camera data\, which are commonly used perce
ption sensors in the mobile robotics. It is important to note that the tim
ealigned discrete signals (3D points and their reflectivity from lidar\, a
nd pixel location and color from camera) of these data streams are generat
ed by sampling the same physical scene\, but in a different manner. Thus\,
although these signals look quite different at a high level (2D image fro
m a camera looks entirely different than a 3D point cloud of the same scen
e from a lidar)\, since they are generated from the same physical scene\,
they are statistically dependent upon each other at signal level. The goal
here is to exploit this statistical dependence\, by taking concepts from
probability and information theory\, and use it in practical robotics appl
ications.\n\nBio: Gaurav Pandey is a PhD candidate in the department of El
ectrical Engineering and Computer Science at the University of Michigan. H
e is currently a Graduate Student Research Assistant (GSRA) at the Percept
ual Robotics Laboratory (http://robots.engin.umich.edu/). His research foc
us is on visual perception for mobile robotics using tools from computer v
ision\, machine learning and information theory. He did his BTech from Ind
ian Institute of Technology\, Roorkee in 2006. Before joining University o
f Michigan for the PhD he worked in Kritikal Solutions Pvt Ltd (KSPL)\, a
student based startup of IIT Delhi\, for two years. In KSPL he worked on v
arious commercial computer vision and image processing related projects.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2179
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2179
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hard-core Set Theorem: Magic of Average Case Complexity
DTSTART;VALUE=DATE-TIME:20120525T093000Z
DTEND;VALUE=DATE-TIME:20120525T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2267@cern.ch
DESCRIPTION:Suppose you are trying to build a circuit using limited number
of AND\, OR\, NOT gates to compute some boolean function f over inputs in
{0\,1}n. But the function is such that every circuit with the above limit
on its resources fails to compute f correctly on a small fraction of the
inputs\, i.e. the function is 'hard' for circuits of this given size.\n\nI
n the above case\, two things might be happening:\n\na) All the circuits t
hat you are trying out are failing for different sets of inputs\, and intu
itively\, there should be functions for which no particular input is harde
r than any particular other input\, per se.\n\nb) There can be one small s
et H hidden in the input set which is extremely hard to compute by any cir
cuit\, and on that set H\, no circuit can do any better than guessing f ra
ndomly.\n\nImpagliazzo's hardcore set lemma is 'one bit of magic in comple
xity theory' which says that it is always the second case\, as if the hard
ness of the function is concentrated on a small 'hardcore' set H inside th
e input space. I will state and prove this result in the talk.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=2267
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2267
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Isolation Lemma
DTSTART;VALUE=DATE-TIME:20120601T093000Z
DTEND;VALUE=DATE-TIME:20120601T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2279@cern.ch
DESCRIPTION:Isolation lemma is a technique used in randomized algorithms t
o reduce the number of solutions of a problem to one\, should a solution e
xist. This is achieved by giving random weights to solutions and showing t
hat there is a unique solution of minimal weight with constant probability
. We will prove this lemma and use it to show that in a computational mode
l called switching networks counting is not weaker than non determinism.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2279
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2279
END:VEVENT
BEGIN:VEVENT
SUMMARY:Probabilistic Systems and Their Verification
DTSTART;VALUE=DATE-TIME:20120608T060000Z
DTEND;VALUE=DATE-TIME:20120608T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2281@cern.ch
DESCRIPTION:The study of automata with probabilistic behaviour began with
Rabin (1963). Using transition systems as models\, specifying their behavi
our using formulas of temporal logic and checking that such a system satis
fies its specification was formulated by Hansson and Jonsson (1994) and by
Bianco and de Alfaro (in FSTTCS 1995). Unlike the key idea on which non-p
robabilistic model checking is based\, emptiness of probabilistic automata
is undecidable\, shown by Paz (1971).\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=2281
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2281
END:VEVENT
BEGIN:VEVENT
SUMMARY:Can We Optimize Path By Path?
DTSTART;VALUE=DATE-TIME:20120608T093000Z
DTEND;VALUE=DATE-TIME:20120608T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2284@cern.ch
DESCRIPTION:Stochastic optimal control problems are usually solved by writ
ing the dynamic programming equation and subsequent evaluation of the asso
ciated value function using various tools\, depending on the context. In t
his talk\, we discuss the possibility of arriving at the value function by
solving a deterministic control problem on individual sample paths. We b
ase our discussion on Rogers[2008]^\, in which the means of doing determin
istic optimization independently on every sample path for solving a gener
al stochastic control problem\, was first established. This amazing connec
tion between stochastic and deterministic optimal control by means of writ
ing out a dual for value function can lead to new computational techniques
.\n\nIn this talk\, we consider a discrete-time Markov process that can be
controlled\; we introduce the optimization problem involved and discuss t
he usual terminologies associated with the control of such a process. Then
\, we proceed to represent the objective function\, in a dual form\, very
similar to how we do in Lagrange multipliers way of solving constrained-op
timization problem. We also discuss the implications of this dual represen
tation on arriving at new computational procedures.\n\n*Nothing more than
familiarity with discrete-time Markov processes is assumed.*\n\nRogers\, L
. C. G. Pathwise stochastic optimal control. SIAM J. Control Optim. 46 (
2007)\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=228
4
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2284
END:VEVENT
BEGIN:VEVENT
SUMMARY:Probabilistic Systems and Their Verification
DTSTART;VALUE=DATE-TIME:20120614T110000Z
DTEND;VALUE=DATE-TIME:20120614T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2282@cern.ch
DESCRIPTION:The study of automata with probabilistic behaviour began with
Rabin (1963). Using transition systems as models\, specifying their behavi
our using formulas of temporal logic and checking that such a system satis
fies its specification was formulated by Hansson and Jonsson (1994) and by
Bianco and de Alfaro (in FSTTCS 1995). Unlike the key idea on which non-p
robabilistic model checking is based\, emptiness of probabilistic automata
is undecidable\, shown by Paz (1971).\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=2282
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2282
END:VEVENT
BEGIN:VEVENT
SUMMARY:Completing Latin Squares
DTSTART;VALUE=DATE-TIME:20120615T093000Z
DTEND;VALUE=DATE-TIME:20120615T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2289@cern.ch
DESCRIPTION:Some of the oldest combinatorial objects\, whose studies appar
ently goes back to ancient times\, are the Latin squares. To obtain a Lati
n square\, one has to fill the cells of an (n x n)- square array with the
numbers 1\,...\,n so that every number appears exactly once in every row a
nd in every column. Say someone started filling the cells with the numbers
{1\,...\,n}. At some point she stops and asks us to fill in the remaining
cells so that we get a Latin square. When is this possible? To find out\,
come to the talk.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=2289
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2289
END:VEVENT
BEGIN:VEVENT
SUMMARY:Formal Design and Analysis of Cyber-Physical Systems
DTSTART;VALUE=DATE-TIME:20120621T060000Z
DTEND;VALUE=DATE-TIME:20120621T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2292@cern.ch
DESCRIPTION:The term ``cyber-physical systems" (CPS) refers to any network
of digital and analog systems whose performance crucially depends on both
the continuous dynamics of the analog parts and the real-time switching
decisions made by the digital system. Such systems are increasingly playin
g safety-critical role in modern life\, where a fault in their design can
be catastrophic. Implantable medical devices are important paradigmatic ex
amples of such safety-critical cyber-physical systems. The growing number
of fatalities due to implantable device failures\, and an ever increasing
list of device recalls by US Food and Drug Administration (FDA) in recent
years underscore the challenges in engineering safety-critical CPS. Simila
r examples can also be cited for CPS from other domains such as avionics\,
automotive\, transportation networks\, and energy sector. The central the
me underlying my research is to employ rigorous mathematical reasoning to
analyze (verification and optimization) and to design (controller synthesi
s) CPS with guaranteed performance. In this talk I will summarize a few ke
y techniques for formal design and analysis of CPS with emphasis upon my r
esearch contributions.\n\nI will also present a recent result on the green
scheduling problem where the goal is to co-ordinate switching decisions o
f various HVAC units so as to minimize the peak energy usage. We pose the
green scheduling problem using optimal scheduling problem for constant-rat
e multi-mode systems (MMS). MMS model hybrid systems that can switch freel
y among a finite set of modes\, and whose dynamics is specified by a finit
e number of real-valued variables with mode-dependent constant rates. The
schedulability problem for such systems is to design a mode-switching poli
cy that maintains the state within a specified safety set. A key result is
that the schedulability problem for MMS can be decided in polynomial tim
e. The talk is based on a joint-work with Prof. Rajeev Alur and Dr. Domini
k Wojtczak which has received the best paper award at 15h conference on hy
brid systems (HSCC'12) at 2012 Cyber-Physical Systems Week.\n\nBio: Ashuto
sh Trivedi is a postdoctoral researcher (since 2011) in the computer and i
nformation science department at the University of Pennsylvania. He works\
, broadly\, in the area of formal analysis and design of real-time and hyb
rid systems. After receiving his PhD from the University of Warwick in 200
9\, he spent 2 years at the University of Oxford as a postdoctoral researc
her working as a part of the Large-Scale Complex IT Systems (LSCITS) proje
ct. Prior to his PhD\, he received a Masters degree in Electrical Engineer
ing (Reliability) from IIT Bombay in 2003\, and has worked as research ass
istant with the CFDVS center at IIT Bombay\, the NuSMV development group
at IRST\, Italy and the formal methods group at University of Tuebingen\,
Germany.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
2292
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2292
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sparsely Supported Correlated Equilibrium
DTSTART;VALUE=DATE-TIME:20120622T093000Z
DTEND;VALUE=DATE-TIME:20120622T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2296@cern.ch
DESCRIPTION:Correlated equilibria are a generalization of Nash equilibria.
Existence of a correlated equilibrium is guaranteed due to existence of a
Nash equilibrium. In this talk I will present a result of Germano & Lugos
i(2007) that shows every finite normal form game possesses a sparsely supp
orted correlated equilibrium.\n\nRef: Fabrizio Germano & Gábor Lugosi\, 2
007. "Existence of Sparsely Supported Correlated Equilibria\," Economic Th
eory\, Springer\, vol. 32(3)\, pages 575-578\, September.\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=2296
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2296
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction to Representation Theory
DTSTART;VALUE=DATE-TIME:20120629T103000Z
DTEND;VALUE=DATE-TIME:20120629T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2309@cern.ch
DESCRIPTION:Representation theory is a branch of mathematics that studies
abstract algebraic structures by representing their elements as linear tra
nsformations of vector spaces. The most prominent of these (and historical
ly the first) is the representation theory of groups. In this talk I will
introduce this subject by mainly focussing on the representation theory of
finite groups. Only knowledge of basic linear algebra and group theory wi
ll be assumed\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=2309
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2309
END:VEVENT
BEGIN:VEVENT
SUMMARY:Correctness of Closed Nesting in Software Transactional Memory Sys
tems
DTSTART;VALUE=DATE-TIME:20120704T060000Z
DTEND;VALUE=DATE-TIME:20120704T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2324@cern.ch
DESCRIPTION:There has been considerable interest in Software Transactional
Memory in recent years. The reason for rise in STM is due to rise of mult
icore computers. To fully utlize the power of these machines\, application
s need to be able to harness the parallelism of the underlying hardware. T
his is commonly achieved using multi-threading. Yet writing correct and sc
alable multi-threaded programs is far from trivial. In multi-threaded prog
rams sets of semantically related actions may need to execute in mutual ex
clusion to avoid semantic inconsistencies.\n\nTraditionally\, multi-thread
ed programs were developed in conjunction with locks to address these issu
es. But programming with locks has many disadvantages such as deadlocks\,
priority inversion etc. and makes it difficult to build scalable software
systems. Importantly\, lock based software components are difficult to com
pose i.e. build larger software systems using simpler software components.
Composition of software components is a very important property which is
the basis of modular programming. Nesting of transactions is a way of achi
eving Composition. \n\nIn this talk\, I will explain about how closed nest
ing can be achieved with Software Transactional Memory. I will talk about
correctness requirements of Closed Nested executions. I will then describe
correctness criteia that we developed and then describe about an implemen
tation based on them.\n\nhttps://indico.tifr.res.in/indico/conferenceDispl
ay.py?confId=2324
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2324
END:VEVENT
BEGIN:VEVENT
SUMMARY:Reconstruction of Binary Functions and Shapes From Incomplete Freq
uency Information
DTSTART;VALUE=DATE-TIME:20120706T093000Z
DTEND;VALUE=DATE-TIME:20120706T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2335@cern.ch
DESCRIPTION:The characterization of a binary function by partial frequency
information is considered. We show that it is possible to reconstruct the
binary signal from incomplete measurements via solving a simple linear op
timization problem. We further prove that if the binary function is spatia
lly structured \, then it can be recovered from very few frequency measure
ments.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=23
35
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2335
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Simple Combinatorial Algorithm for Submodular Function Minimizat
ion
DTSTART;VALUE=DATE-TIME:20120713T093000Z
DTEND;VALUE=DATE-TIME:20120713T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2343@cern.ch
DESCRIPTION:Submodular functions are important to study as they arise in m
any context as flow problems\, game theoretic application etc. One importa
nt aspect to study about submodular function is the submodular function mi
nimization. I will present an algorithm for minimizing integer valued subm
odular function. The algorithm runs in O(n6.EO.lognM) time\, where n is ca
rdinality of ground set\, M is maximum absolute value of the function\, an
d EO is the time for function evaluation.\n\nRef: Iwata S\, Orlin J\; A Si
mple Combinatorial Algorithm for Submodular Function Minimization\; SODA 2
009\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2343
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2343
END:VEVENT
BEGIN:VEVENT
SUMMARY:Recovering from Adversarial Error in Boolean Circuits
DTSTART;VALUE=DATE-TIME:20120718T103000Z
DTEND;VALUE=DATE-TIME:20120718T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2352@cern.ch
DESCRIPTION:We consider two different models for adversarial errors and sh
ow how to design circuits that can recover from such errors. \n\nIn the fi
rst work\, we show how to efficiently convert any boolean formula F into a
boolean formula E that is resilient to short-circuit errors (as introduce
d by Kleitman et al.). A gate has a short-circuit error when the value it
computes is replaced by the value of one of its inputs.\n\nWe guarantee th
at E computes the same function as F\, as long as at most (1/10−ϵ) of t
he gates on each path from the output to an input have been corrupted in E
. The corruptions may be chosen adversarially\, and may depend on the form
ula E and even on the input. We obtain our result by extending the Karchme
r-Wigderson connection between formulas and communication protocols to the
setting of adversarial error. This enables us to obtain error-resilient f
ormulas from error-resilient communication protocols.\n\nIn the second wor
k\, we assume that the input has been encoded using a suitable error corre
cting code C. We show that given any boolean circuit F\, one can design a
layered circuit E such that E applied to the encoded input produces an out
put that is close to the error corrected version of the output of F\, as l
ong as at most a constant fraction of the gates in each layer of the circu
it are corrupted (the corruptions may be arbitrary).\n\nThe first result i
s joint work with Yael Kalai and Allison Lewko. The second is joint work w
ith Aram Harrow and Matthew Hastings.\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=2352
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2352
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Sunflower Lemma and Its Application to Circuit Lower Bound
DTSTART;VALUE=DATE-TIME:20120720T093000Z
DTEND;VALUE=DATE-TIME:20120720T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2360@cern.ch
DESCRIPTION:Sunflowers are highly regular configurations in extremal set t
heory. The sunflower lemma discovered by Erdos and Rado in 1960 asserts th
at in a sufficiently large uniform family\, some highly regular configurat
ion called "Sunflower" must occur\, regardless of the size of the universe
. In this talk we'll consider this result as well as one of its modificati
on (called Flower lemma) and its application. \n\nThe sunflower lemma and
its modifications have many applications in computational complexity theor
y. We'll only prove lower bound of a special 3-depth formula computing an
s-threshold function (An s-threshold function is a monotone Boolean functi
on which outputs 1 iff at least s of its inputs are 1).\n\nReference : Cha
pter 6\, Extremal Combinatorics with applications in computer science\, 2n
d edition\, Springer (Author : Stasys Jukna)\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=2360
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2360
END:VEVENT
BEGIN:VEVENT
SUMMARY:Findinding Top k Elements and Heavy Hitters in Data Streams
DTSTART;VALUE=DATE-TIME:20120727T093000Z
DTEND;VALUE=DATE-TIME:20120727T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2368@cern.ch
DESCRIPTION:Finding top k items and finding heavy hitters in data streams
are two well studied problems in streaming algorithms. We shall discuss t
wo deterministic streaming algorithms\, one for each of these two problems
.\n\nReference: Ahmed Metwally\, Divyakant Agrawal\, Amr El Abbadi: Effici
ent Computation of Frequent and Top-k Elements in Data Streams. ICDT 2005.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2368
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2368
END:VEVENT
BEGIN:VEVENT
SUMMARY:Linear-algebraic List Decoding and Subspace-evasive Sets
DTSTART;VALUE=DATE-TIME:20120731T103000Z
DTEND;VALUE=DATE-TIME:20120731T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2370@cern.ch
DESCRIPTION:We will describe a simple linear-algebraic approach to list de
code Reed-Solomon codes with evaluation points from a subfield. The algori
thm is able to correct an error fraction approaching the information-theor
etically maximum limit of 1−R where R is the rate of the code.\n\nThe al
gorithm can be thought of as a higher-dimensional version of the Welch-Ber
lekamp decoder\, and pins down the candidate solutions to a subspace of no
n-trivially smaller dimension. By pre-coding the messages to belong to a p
seudorandomly constructed "subspace-evasive set" that has small intersecti
on with subspaces of the sort output by the decoder\, one can prune the su
bspace to a small list of close-by codewords in polynomial time.\n\nThis a
pproach to list decoding is quite versatile\, and applies to variants of R
eed-Solomon codes (the context where it was first discovered)\, folded alg
ebraic-geometric codes (which yields efficiently list-decodable codes with
simultaneously near-optimal rate\, list size\, and alphabet size)\, and G
abidulin codes (the rank-metric analog of Reed-Solomon codes). Time permit
ting\, we might briefly touch upon some of these variants.\n\nBased on joi
nt work with Chaoping Xing.\n\nhttps://indico.tifr.res.in/indico/conferenc
eDisplay.py?confId=2370
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2370
END:VEVENT
BEGIN:VEVENT
SUMMARY:Broadcasting Private Message Securely
DTSTART;VALUE=DATE-TIME:20120803T093000Z
DTEND;VALUE=DATE-TIME:20120803T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2384@cern.ch
DESCRIPTION:Wireless communication channels are easier to eavesdrop and ha
rder to secure – even towards unintentional eavesdrop- pers. As an examp
le consider a sender\, Alice\, who wants to send private messages to multi
ple (say three) receivers\, Bob\, Calvin and David\, within her transmissi
on radius\, and assume public feedback from the receivers to Alice. When A
lice broadcasts a message W1 intended for Bob\, Calvin and David should al
so try to overhear\, as the side information they possibly collect can ena
ble Alice to make her following broadcast transmissions more efficient\; b
ut then\, this collected side information would allow Calvin and David to
learn parts of Bob’s message. Even worse\, Calvin and David could try to
put together the parts they overheard\, to extract increased information
about Bob’s message. Can we\, in such a setting\, keep the message for e
ach user information theoretically secure from the other users\, even if t
hese collaborate? Moreover\, can we do so\, when the users can only commun
icate through shared wireless broadcast channels?\n\nhttps://indico.tifr.r
es.in/indico/conferenceDisplay.py?confId=2384
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2384
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Strong Converses for the Quantum Channel Capacity Theorems
DTSTART;VALUE=DATE-TIME:20120807T093000Z
DTEND;VALUE=DATE-TIME:20120807T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2389@cern.ch
DESCRIPTION:A unified approach to prove the converses for the quantum chan
nel capacity theorems is presented. These converses include the strong con
verse theorems for classical or quantum information transfer with error ex
ponents and novel explicit upper bounds on the fidelity measures reminisce
nt of the Wolfowitz strong converse for the classical channel capacity the
orems. We provide a new proof for the error exponents for the classical in
formation transfer. A long standing problem in quantum information theory
has been to find out the strong converse for the channel capacity theorem
when quantum information is sent across the channel. We give the quantum e
rror exponent thereby giving a one-shot exponential upper bound on the fid
elity. We then apply our results to show that the strong converse holds fo
r the quantum information transfer across an erasure channel for maximally
entangled channel inputs (joint work with Naqueeb A. Warsi).\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=2389
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2389
END:VEVENT
BEGIN:VEVENT
SUMMARY:Cognitive Computing and the Network Architecture of the Long Dista
nce Pathways in the Macaque Brain
DTSTART;VALUE=DATE-TIME:20120810T090000Z
DTEND;VALUE=DATE-TIME:20120810T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2379@cern.ch
DESCRIPTION:Cognitive computing aims to develop a coherent\, unified\, uni
versal mechanism inspired by the mind’s capabilities. Rather than assemb
le a collection of piecemeal solutions\, whereby different cognitive proce
sses are each constructed via independent solutions\, we seek to implement
a unified computational theory of the mind.\n\nNeuroanatomy places critic
al constraints on the functional connectivity of the cerebral cortex. To
understand and model these constraints\, the brain network has to be extra
cted and and its properties analyzed. We have derived a unique network inc
orporating 410 anatomical tracing studies of the macaque brain. The networ
k consists of 383 hierarchically organized regions spanning cortex\, thala
mus\, and basal ganglia and models the presence of 6\,602 directed long-di
stance connections.\n\nImportant network-theoretic properties that we foun
d include: the brain has a exponential degree distribution\; the pre-front
al cortex\, the region implicated in executive functions\, is central unde
r many different topological centrality measures\; and there exists a tigh
ly integrated core in the brain network. In this talk I will discuss the
brain network\, its properties\, and how this is useful in the cognitive c
omputing paradigm.\n\nBio - Raghavendra Singh is a Research Staff member a
t IBM Research - India (Delhi). He is currently a member of the Cognitive
Computing group. His research interests are in the area of information the
ory\, signal processing and representation as applied to a spectrum of pro
blems in neuro-science\, compression and transmission of multimedia data\,
and monitoring of large-scale data centers.\n\nHe did his PhD. in Electri
cal Engineering from University of Southern California (2001). His undergr
aduate degree is from BITS Pilani\, India (1993). He is a Senior Member of
the IEEE.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=2379
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2379
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Bounds for the Lambert W Function and its Application
DTSTART;VALUE=DATE-TIME:20120817T093000Z
DTEND;VALUE=DATE-TIME:20120817T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2404@cern.ch
DESCRIPTION:Lambert function is defined as the multivalued inverse relatio
n of the function f(w)=wexp(w). We will study some of its important proper
ties and prove some bounds. We will also look at some simple applications
of the Lambert W function in different domains.\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=2404
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2404
END:VEVENT
BEGIN:VEVENT
SUMMARY:Load Balancing Via Random Local Search
DTSTART;VALUE=DATE-TIME:20120823T053000Z
DTEND;VALUE=DATE-TIME:20120823T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2408@cern.ch
DESCRIPTION:We analyze the performance of random load resampling and migra
tion strategies in parallel server systems. Clients initially attach to an
arbitrary server\, but may switch server independently at random instants
of time in an attempt to improve their service rate. This approach to loa
d balancing contrasts with traditional approaches where clients make smart
server selections upon arrival (e.g.\, Join-the-Shortest-Queue policy and
variants thereof). Load resampling is particularly relevant in scenarios
where clients cannot predict the load of a server before being actually at
tached to it. An important example is in wireless spectrum sharing where c
lients try to share a set of frequency bands in a distributed manner.\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2408
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2408
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Online Network Tomography Algorithm
DTSTART;VALUE=DATE-TIME:20120824T093000Z
DTEND;VALUE=DATE-TIME:20120824T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2419@cern.ch
DESCRIPTION:Network tomography is the science of inferring spatially local
ized network behavior using only metrics that are practically feasible to
measure. Mathematically\, given a matrix A\, the goal of network tomograph
y is to estimate the statistics of X\, a vector of mutually independent ra
ndom variables\, from the measurement model Y=AX. The challenge in these p
roblems stems from the fact that A is usually an ill-posed matrix and henc
e non-invertible. In this talk\, using the stochastic approximation varian
t of the Kaczmarz's (SAK) algortihm\, we will see an online scheme for est
imating the expected value of X using only IID samples of the components o
f the random vector Y. Importantly\, we will prove that\, starting from t
he same initial point\, the SAK algorithm\, when only samples of component
s of Y are available\, and the usual Kaczmarz algorithm\, when EY is exact
ly known\, converge to precisely the same point (this result is joint work
with Prof. Vivek Borkar and Prof. D. Manjunath (IIT Mumbai)).\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=2419
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2419
END:VEVENT
BEGIN:VEVENT
SUMMARY:State-independent Importance Sampling for Regularly Varying Random
Walks
DTSTART;VALUE=DATE-TIME:20120831T053000Z
DTEND;VALUE=DATE-TIME:20120831T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2432@cern.ch
DESCRIPTION:Efficient simulation of rare events involving sums of heavy-ta
iled random variables has been an active research area in applied probabil
ity in the last fifteen years. These rare events arise in many application
s including telecommunications\, computer and communication networks\, ins
urance and finance. These problems are viewed as challenging\, since large
deviations theory inspired and exponential twisting based importance samp
ling algorithms that work well for rare events involving sums of light tai
led random variables fail in these settings.\n\nIn this talk we shall disc
uss about developing some simple state-independent exponential twisting ba
sed importance sampling methods to efficiently estimate such rare event pr
obabilities. Specifically\, we develop strongly efficient algorithms for e
stimating:\n\n1. The classical large deviations probability that the sums
of independent\, identically distributed random variables with regularly v
arying tails exceed an increasing threshold both in the case where the num
ber of random variables increases to infinity and when it is fixed.\n\n2.
Finite-horizon level crossing probabilities for negative-mean regularly va
rying random walks\n\nAccurate computation of these level crossing probabi
lities has applications in estimating ruin probabilities in insurance sett
ings and calculating waiting times in GI/G/1 queues.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=2432
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2432
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Multiplicative Weights Method and Application to Solving LPs
DTSTART;VALUE=DATE-TIME:20120831T093000Z
DTEND;VALUE=DATE-TIME:20120831T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2434@cern.ch
DESCRIPTION:Consider n 'experts' predicting the outcome of\, say\, the sto
ck market\, with errors. At the start of every day\, they make a predictio
n of whether the market is going up or down\, and then the actual outcome
is revealed to us at the end of the day. We are then free to choose the ex
perts we trust for the next day. The Multiplicative Weights Method is a si
mple yet powerful method that gives us a way to assign importance to these
experts at every round so that over a period of ln n days\, the fraction
of errors we make is close to that of the best expert.\n\nSurprisingly\, t
his method has been used to solve certain kinds of LP relaxations of combi
natorial optimization problems faster than that of a general LP solver. Fu
rther\, it has been generalized to SDPs and SDP-bsed algorithms as well.\n
\nWe shall see a proof of correctness of the Multiplicative Weights method
\, and the application to solving LPs.\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=2434
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2434
END:VEVENT
BEGIN:VEVENT
SUMMARY:Competitive Equilibria with Budget Limits
DTSTART;VALUE=DATE-TIME:20120905T103000Z
DTEND;VALUE=DATE-TIME:20120905T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2428@cern.ch
DESCRIPTION:We consider the problem of finding a competitive equilibrium w
hen agents have budget constraints and items are indivisible. In our model
\, the utility of an agent is equal to her valuation minus payment when th
e payment is no more than her budget\; otherwise she gets negative utility
. In a competitive equilibrium\, every item is assigned a non-negative pri
ce\, and the items are allocated to the agents in such a way that every ag
ent gets her utility-maximizing bundle\, and every unallocated item has pr
ice zero.\n\nMost previous work on competitive equilibria (with and withou
t budgets) has focused on the {\\em Gross Substitutes} case\, where a comp
etitive equilibrium is guaranteed to exist modulo tie-breaking. Our work d
eals with two types of valuation functions that do not satisfy this proper
ty in the presence of budget constraints: Additive valuations\, and concav
e combinatorial valuations.\n\nWe present some hardness results for decidi
ng the existence of exact competitive equilibria. Next\, we develop connec
tions to a related Fisher model of market clearing in order to design poly
-time algorithms that compute approximate competitive equilibria. Not only
do our algorithms require several new technical ideas\, but we also show
that this approach sheds new light on algorithms for the Fisher model. In
particular\, it yields an interpretation of the well-known Eisenberg-Gale
convex program as optimizing a measure of social welfare. We also prove
strong revenue properties of the equilibria we construct. We thereby impr
ove and extend the recent work on revenue maximizing envy-free multi-unit
auctions by Feldman et al. in EC 2012 (joint work with Vincent Conitzer\,
Kamesh Munagala and Xiaoming Xu).\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=2428
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2428
END:VEVENT
BEGIN:VEVENT
SUMMARY:Improved Prior-Free Auctions with Ordered Bidders
DTSTART;VALUE=DATE-TIME:20120906T083000Z
DTEND;VALUE=DATE-TIME:20120906T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2429@cern.ch
DESCRIPTION:A central problem in Microeconomics is to design auctions with
good revenue properties. Consider the following setting. Multiple bidders
are participating in an auction. The bidders' valuations for the items ar
e private knowledge\, but they are drawn from publicly known prior distrib
utions. The goal is to implement a truthful auction (no bidder can gain in
utility by misreporting her valuation) that maximizes the expected revenu
e. Here\, the expectation is over the prior distributions of the bidders'
valuations\, and the random choices made by the auctioneer.\n\nNaturally\,
to execute the auction with optimal expected revenue\, the auctioneer fir
st needs to know the prior distributions. An intriguing question is to des
ign a truthful auction that is oblivious to these priors\, and yet manages
to get a constant factor of the optimal revenue. Such auctions are called
{\\em prior-free}. This question was posed in a seminal paper by Goldberg
et al. [SODA 2001].\n\nGoldberg et al. presented a constant approximate p
rior-free auction when there are identical copies of an item available in
unlimited supply\, bidders are unit-demand\, and their valuations are draw
n from i.i.d. distributions. Very recently\, Leonardi et al. [STOC 2012]
generalized this setting to non i.i.d. bidders when the auctioneer knows t
he ordering of their reserve prices. Leonardi et al. presented a prior-fre
e auction that achieves a O(log∗n) approximation. We improve upon this r
esult and give a simple prior-free auction with constant approximation gua
rantee (joint work with Janardhan Kulkarni\, Kamesh Munagala and Xiaoming
Xu).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2429
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2429
END:VEVENT
BEGIN:VEVENT
SUMMARY:Size-energy Tradeoffs for Unate Circuits Computing Symmetric Boole
an Functions
DTSTART;VALUE=DATE-TIME:20120907T093000Z
DTEND;VALUE=DATE-TIME:20120907T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2448@cern.ch
DESCRIPTION:A unate gate is a logical gate computing a unate Boolean funct
ion\, which is monotone in each variable. Examples of unate gates are AND
gates\, OR gates\, NOT gates\, threshold gates etc. A unate circuit C is a
combinatorial logic circuit consisting of unate gates. Let f be a symmetr
ic Boolean function of n variables. Let m0 and m1 be the maximum number of
consecutive 0s and consecutive 1s in the value vector v(f) of f\, respect
ively. Also\, let l=minm0\,m1 and m=maxm0\,m1. Now\, let C be a unate circ
uit computing f. Let s be the size of the circuit C\, that is\, C consists
of s unate gates. If we define the maximum number of gates outputting '1'
over all inputs to C to be the energy e of the circuit C\, then we can sh
ow that there exists a tradeoff between the size s and the energy e of C.
More precisely\, we will show the following - (n+1−l)/m\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=2448
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2448
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Minimal Complexity of an Adapting Agent Increases with Fitness
DTSTART;VALUE=DATE-TIME:20120913T110000Z
DTEND;VALUE=DATE-TIME:20120913T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2397@cern.ch
DESCRIPTION:What is the relationship between the complexity and the fitnes
s of evolved organisms\, whether natural or artificial? It has been assert
ed\, primarily based on empirical evidence\, that the complexity of biolog
ical organisms increases as their fitness within a particular environment
increases via evolution by natural selection. In this work\, we derive an
analytical expression for the minimal mutual information I(s:m) quantifyi
ng the organizational correlation between the sensory input s and the corr
esponding motor response m required for achieving a given fitness level F
in a given environment . We show that under certain conditions\, the minim
al mutual information I(s:m) is a semi-positive definite\, monotonic non-d
ecreasing\, concave function of F.\n\nWe also simulated the adaptation of
brains of digital organisms living in static mazes\, whose {\\it connectom
e} evolves over 10\,000s of generations from a random initial state. We c
omputed the circuit complexity of these brains as they evolve to adapt to
their environment using an entropy-based measure known as {\\it integrated
information}. In a confirmation of our theoretical derivation\, we find t
hat the minimal complexity of the animats' brains increases with their fit
ness. That is\, to achieve any given level of fitness\, there exists a min
imal amount of circuit complexity.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=2397
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2397
END:VEVENT
BEGIN:VEVENT
SUMMARY:Electrical Networks and Spanning Trees
DTSTART;VALUE=DATE-TIME:20120914T093000Z
DTEND;VALUE=DATE-TIME:20120914T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2467@cern.ch
DESCRIPTION:We will go through some connections between current flow in an
electrical network and the number of spanning trees in the underlying gra
ph.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2467
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2467
END:VEVENT
BEGIN:VEVENT
SUMMARY:Non i.i.d. Information Theory
DTSTART;VALUE=DATE-TIME:20120921T090000Z
DTEND;VALUE=DATE-TIME:20120921T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2474@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
2474
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2474
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quadratic Dynamical Systems
DTSTART;VALUE=DATE-TIME:20120928T093000Z
DTEND;VALUE=DATE-TIME:20120928T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2481@cern.ch
DESCRIPTION:The main purpose of this talk will be to promote the study of
computational aspects\, primarily the convergence rate\, of non-linear dyn
amical systems from a combinatorial perspective.\n\nMany natural phenomeno
n can be described by dynamical systems in Euclidean space. In such a desc
ription\, there is a fixed set of types\, and a point in the system (usual
ly referred to as a 'state' or a 'population') specifies how many elements
of each type exist at a given time instant. The system evolves under a fi
xed map f\, which attempts to capture the underlying phenomenon. A couple
of examples follow that help illustrate the applicability of these systems
.\n\nIn Physics\, the system might describe the behaviour of gas molecules
in a container. In this case\, the types correspond to velocity values\,
and the state p specifies how many molecules of each type there are at a c
ertain point in time. The map f should include the Newtonian laws so as to
produce from each state p a new state f(p) in the next time step\, under
some assumption about the spatial distribution of the molecules.\n\nAgain\
, in Biology\, the types may correspond to the genotypes of some species.
A population p is then simply the number of individuals of each type in th
e current generation. The map f determines the population in the next gene
ration according to a fixed set of rules that includes the genetic outcome
of mating\, the survival capacity of different types\, random mutations e
tc. Thus\, such systems also provide an appropriate framework for the stud
y of genetic algorithms in combinatorial optimization.\n\nIn this talk\, w
e will first identify the class of symmetric quadratic dynamical systems.
Then\, we will go on to prove several fundamental general properties of th
ese systems\, including a characterization of the set of fixed points to w
hich the system converges.\n\nThis talk will be based on the preliminary v
ersion of a paper titled "Quadratic Dynamical Systems" authored by Yuri Ra
binovich\, Alistair Sinclair and Avi Wigderson.\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=2481
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2481
END:VEVENT
BEGIN:VEVENT
SUMMARY:Multilevel Monte Carlo Path Simulation
DTSTART;VALUE=DATE-TIME:20121005T093000Z
DTEND;VALUE=DATE-TIME:20121005T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2505@cern.ch
DESCRIPTION:We will see how multigrid ideas can be used to reduce the comp
utational complexity (computational cost) of estimating an expected value
arising from the solution of a stochastic differential equation using Mont
e Carlo path simulations. In the simplest case of a Lipschitz payoff and a
Euler discretization\, the computational cost to achieve an accuracy of O
(e) is reduced from O(e^-3) to O(e^-2 * (log e)^2). A brief overview of re
lated concepts will be provided before discussing the main details of the
paper. (e = \\epsilon)\n\n*Reference: *Giles\, M.B.\, Multilevel Monte Car
lo Path Simulation\, Operations Research\, 2008.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=2505
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2505
END:VEVENT
BEGIN:VEVENT
SUMMARY:Balanced Information Inequalities
DTSTART;VALUE=DATE-TIME:20121009T100000Z
DTEND;VALUE=DATE-TIME:20121009T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2503@cern.ch
DESCRIPTION:In this talk we will discuss about linear information inequali
ties\, both discrete and continuous ones. We will prove that every discret
e information inequality is associated with a “balanced” information i
nequality and a set of “residual weights.” To prove the inequality\, i
t is necessary and sufficient to prove that its “balanced” version is
valid and all its residual weights are nonnegative. For a continuous infor
mation inequality\, we will prove that it is valid if and only if its disc
rete counterpart is balanced and valid.\n\nThis work is due to Terence H.
Chan.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=250
3
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2503
END:VEVENT
BEGIN:VEVENT
SUMMARY:Proof Transformations for Verification
DTSTART;VALUE=DATE-TIME:20121012T100000Z
DTEND;VALUE=DATE-TIME:20121012T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2490@cern.ch
DESCRIPTION:Various verification methods depend on theorem provers to obta
in proofs of verification conditions. If the provers return proofs that sa
tisfy certain structure then it may enhance the performance of the verific
ation methods. Theorem provers aim to find the proofs using the most effic
ient algorithms. Therefore\, the returned proofs from the existing theorem
provers may not satisfy such structures. In my talk\, I will present two
pieces of our work that transform the proofs produced by the provers to ob
tain such structured proofs.\n\nIn the first work\, we present a new effic
ient parametrized method to remove redundant parts of resolution proofs. T
he parameter allows one to choose between coverage of the redundancy remov
al and the efficiency of the method.\n\nIn the second work\, we present a
set of proof transformation rules that obtains "localize proofs" from proo
fs of unsatisfiable formulas. The localize proofs allow one to compute (tr
ee)-interpolants efficiently.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=2490
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2490
END:VEVENT
BEGIN:VEVENT
SUMMARY:Pseudorandomness and Unpredictability
DTSTART;VALUE=DATE-TIME:20121012T083000Z
DTEND;VALUE=DATE-TIME:20121012T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2524@cern.ch
DESCRIPTION:This talk will be an introduction to pseudorandomness. We will
motivate it's study and connect it to 'unpredictability' through a theore
m by Yao.\n\nReference: Computational Complexity\, Arora and Barak\, chapt
er 20 (Derandomization)\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=2524
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2524
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Random Threshold Graph Model
DTSTART;VALUE=DATE-TIME:20121019T043000Z
DTEND;VALUE=DATE-TIME:20121019T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2519@cern.ch
DESCRIPTION:Let G be a random graph generated as follows:- each vertex i o
f the vertex set {1\,…\,n} has an associated random variable Xi where {X
i:i≥1} are i.i.d. Two vertices i and j are connected by an edge if and o
nly if\, Xi+Xj>θ where θ is a given threshold value. We study some prope
rties of this graph G.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=2519
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2519
END:VEVENT
BEGIN:VEVENT
SUMMARY:Five-Coloring Plane Graphs
DTSTART;VALUE=DATE-TIME:20121019T093000Z
DTEND;VALUE=DATE-TIME:20121019T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2538@cern.ch
DESCRIPTION:The four-color theorem states that it is always possible to co
lor the regions of a plane map with four colors such that regions that sha
re a boundary receive different colors. This theorem was proven in 1976 by
Appel and Haken. The proof of the theorem required the calculating power
of computers. Although I have a good computer (probably more powerful than
the computer that was used in the original proof)\, we will not try to pr
ove the four-color theorem. We will be modest and prove that we can color
a plane map with five colors.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=2538
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2538
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimizing Pathwise
DTSTART;VALUE=DATE-TIME:20121102T093000Z
DTEND;VALUE=DATE-TIME:20121102T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2556@cern.ch
DESCRIPTION:In this talk\, I'll discuss the pathwise optimization (PO) met
hod for stochastic control problem. We will first see how the method produ
ces upper and lower bounds on the optimal value of a high-dimensional opti
mal stopping problem. The PO methods builds on a dual characterization of
optimal stopping problems as optimization problems over the space of marti
ngales. In the second part of the talk\, I'll discuss how we can extend th
e PO method to solve general stochastic control problem.\n\nReference: Pat
hwise optimization for optimal stopping problems by V. Desai\, V. Farias\,
C. Moallemi\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=2556
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2556
END:VEVENT
BEGIN:VEVENT
SUMMARY:Networks That Fix Themselves Aka Self-healing Networks
DTSTART;VALUE=DATE-TIME:20121106T103000Z
DTEND;VALUE=DATE-TIME:20121106T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2547@cern.ch
DESCRIPTION:Given a connected graph\, two players play a turn-based game:
First. the red guy removes a node (and therefore\, its adjoining edges too
)\, now the blue guy adds edges between the remaining nodes. What edges sh
ould the blue guy add so that over a whole run of the game\, the network r
emains connected\, no node gets too many new edges and the distance betwee
n any pair of nodes (i.e. the network stretch) does not blow up by much? N
ow\, imagine that the nodes in the graph are computers and the graph is a
distributed network\; the nodes themselves are the blue guys but they do n
ot know anybody beyond the nodes they share an edge with. Solving such pro
blems is the essence of self-healing distributed networks.\n\nWe shall pre
sent the distributed self-healing model which is especially applicable to
reconfigurable networks such as peer-to-peer and wireless mesh networks an
d present fully distributed algorithms that can 'heal' certain global and
topological properties using only local information. Forgiving Graph [PODC
2009\; DC 2012] uses a 'virtual graphs' approach maintaining connectivity
\, low degree increase and closeness of nodes (i.e. stretch). Xheal [PODC
2011\; Xheal: localized self-healing using expanders] further maintains ex
pansion and spectral properties of the network. We present a full distribu
ted implementation in the LOCAL message passing model. However\, we are wo
rking on ideas to allow even more efficient implementations and stronger g
uarantees.\n\nBrief bio:\n\nAmitabh Trehan is a postdoc with the informati
on systems group at the faculty of Industrial Engineering at Technion\, Ha
ifa\, Israel. At Technion\, he works with Profs. Shay Kutten and Ron Lavi
on distributed systems and game theory. Before Technion\, he was a postdoc
with Prof. Valerie King (at UVIC\, Canada). His Ph.D. Advisor was Prof.
Jared Saia (University of New Mexico\, USA) with whom he worked on algorit
hms for self-healing networks. His broad research interests are in theory
and algorithms with specific interests in distributed algorithms\, network
s\, and game theory. His interest includes designing efficient distributed
algorithms for robustness/self-healing/self-* properties in systems under
adversarial attack\, and studying game theoretic and other mechanisms for
evolving networks\, such as social networks or distributed systems (P2P n
etworks etc).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=2547
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2547
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Lower Bound for Additive Spanners
DTSTART;VALUE=DATE-TIME:20121109T093000Z
DTEND;VALUE=DATE-TIME:20121109T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2564@cern.ch
DESCRIPTION:Given an undirected unweighted graph G\, a \\beta-additive spa
nner of G is a subgraph H of G in which the shortest distance between any
pair of vertices is stretched within an additive factor \\beta of their sh
ortest distance in G. The stretch of a subgraph H of G is defined as the s
tretch of the worst stretched pair in H. It is clear that\, the stretch in
creases as the subgraph gets sparser. In this talk\, we'll discuss a lower
bound on the number of edges in a spanner having a fixed stretch.\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2564
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2564
END:VEVENT
BEGIN:VEVENT
SUMMARY:Patterning the Embryo: Scaling Pattern With Size During Embryonic
Development
DTSTART;VALUE=DATE-TIME:20121116T090000Z
DTEND;VALUE=DATE-TIME:20121116T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2562@cern.ch
DESCRIPTION:Individuals of the same or closely related species can vary su
bstantially in size. However\, the proportions within (and between) tissue
s are precisely kept. This adaptation of pattern with size\, termed scalin
g\, is receiving a growing attention in recent years. I will discuss exper
imental evidence for scaling during the early development of multicellular
organisms\, and describe theoretical ideas of what is required for scalin
g morphogen gradients. In particular\, I will discuss the Expansion-Repres
sion mechanism which we recently describe and present experimental evidenc
e for its function in two biological systems: the early frog embryo and th
e fruit-fly wing disc.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=2562
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2562
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rate Distortion Theory for Arbitrary Sources
DTSTART;VALUE=DATE-TIME:20121123T093000Z
DTEND;VALUE=DATE-TIME:20121123T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2603@cern.ch
DESCRIPTION:In this talk we will discuss the rate distortion theory from t
he viewpoint of the information-spectrum so that we can treat arbitrary ge
neral sources (not necessarily memoryless and stationary)·\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=2603
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2603
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Shannon Capacity of the 5-Cycle
DTSTART;VALUE=DATE-TIME:20121130T093000Z
DTEND;VALUE=DATE-TIME:20121130T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2624@cern.ch
DESCRIPTION:Imagine a secret agent in a terrorist camp who needs to get me
ssages to the outside world. He has scarves of 5 colours (A\,B\,C\,D\,E)\,
and depending on the colour of the scarf he wears each day\, an intellige
nce team who have satellite photography of the area decipher the message.\
n\nUnfortunately\, the satellite system is not perfect and tends to mix up
certain colours in it's images: A may be confused with E or B\, B with A
or C\, and so on. The confusion graph forms a 5-cycle A-B-C-D-E-A. What pr
otocol can the agent use to transmit his messages with the highest efficie
ncy? (in terms of #messages per symbol)\n\nFor instance\, if he uses just
colours A\,C\, then he transmits 1 bit every day and these are not confuse
d. So he sends one of 2^k messages in k days. But there is a better protoc
ol that allows him to transfer one of 5^(k/2) messages in k days.\n\nIn a
gem of a proof\, Lovasz showed that this is indeed tight\, and the agent c
an do no better. We shall see the proof of this result in the talk.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2624
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2624
END:VEVENT
BEGIN:VEVENT
SUMMARY:Compgen Initiative: New Interfaces Between Computing and Genomics
DTSTART;VALUE=DATE-TIME:20121205T090000Z
DTEND;VALUE=DATE-TIME:20121205T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2630@cern.ch
DESCRIPTION:Given the rate of innovation in gene sequencing technologies\,
through which the cost of sequencing a genome has fallen exponentially fo
r decades\, the amount of genomic data is quickly becoming difficult to ma
nage and index. Although the genomes of two members of a species or the ge
nomes of two closely-related species share significant structure\, much of
the science necessary to identify and understand the implications of this
structure has yet to be done\, in part because of the relative unavailabi
lity\, until quite recently\, of large data sets. As a result\, many algor
ithmic approaches rely on simpler comparisons of genetic sequences or geno
mes and optimization techniques that may not scale to the massive data set
s that are already becoming common. Indeed\, current algorithms and the un
derlying hardware and software architectures are increasingly becoming the
limiting factors. The large volume of genomic data now available provides
an opportunity to identify the structural relationships that will support
algorithmic innovations and configurable hardware extensions to enable sc
alable comparisons with a broad range of existing data. This opportunity
could lead to major advances in scientific understanding through a combina
tion of improved statistical models\, algorithmic advances\, efficient par
allel computation\, and genomic data management. This talk will outline
some of the opportunities for collaboration being investigated by an inter
-disciplinary group of researchers centered at the Institute of Genomic Bi
ology working in close collaboration with computer engineers and scientist
s in the departments of Electrical and Computer Engineering and Computer S
cience at Illinois. \n\nBio:\n\nProfessor Gene Robinson is the Swanlund Ch
air Professor and the Director of the Institute for Genomic Biology with
appointments in the department of Entomology and the Beckman Institute at
the University of Illinois at Urbana Champaign.\n\nProfessor Ravishankar K
. Iyer is the George and Ann Fisher Distinguished Professor of Engineering
with appointments is the Department of Electrical ad Computer Engineering
\, Computers Science and the Coordinated Science Laboratory at the Univers
ity of Illinois at Urbana Champaign.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=2630
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2630
END:VEVENT
BEGIN:VEVENT
SUMMARY:Convergent Sequences of Dense Graphs
DTSTART;VALUE=DATE-TIME:20121207T093000Z
DTEND;VALUE=DATE-TIME:20121207T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2636@cern.ch
DESCRIPTION:In this talk\, we shall discuss various notions of convergence
of dense graphs\, and attempt to come up with a metric space where the no
tion of "left convergence" of graphs can be identified with Cauchy converg
ence. This will be an attempt to get a glimpse on the recent line of work
on convergence in graphs by Lovász and his co-authors.\n\nReferences:\nBo
rgs\, C.\, Chayes\, J. T.\, Lovász\, L.\, Sós\, V. T.\, & Vesztergombi\,
K. 2008). Convergent sequences of dense graphs I: Subgraph frequencies\,
metric properties and testing. Advances in Mathematics\, 219(6)\, 1801-18
51.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2636
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2636
END:VEVENT
BEGIN:VEVENT
SUMMARY:AdaBoost - New Boosting Algorithm
DTSTART;VALUE=DATE-TIME:20121221T093000Z
DTEND;VALUE=DATE-TIME:20121221T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2661@cern.ch
DESCRIPTION:Boosting is a general method to improve the accuracy of any gi
venlearning algorithm. I will introduce a new method of boosting - AdaBoos
t- and discuss its advantages compared to other traditional methods of boo
sting like "boost-by-majority" algorithm. I will also try to give an intui
tive explanation as to why boosting does not suffer from overfitting.\n\nR
eference:\n\nhttp://link.springer.com/chapter/10.1007%2F3-540-59119-2_166?
LI=true\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2
661
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2661
END:VEVENT
BEGIN:VEVENT
SUMMARY:Epistemic Reasoning and Its Applications
DTSTART;VALUE=DATE-TIME:20121231T053000Z
DTEND;VALUE=DATE-TIME:20121231T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2654@cern.ch
DESCRIPTION:It is clear from examples from Shakespeare and others that kno
wledge and belief influence behavior.\n\nBut to understand these examples
some technical tools are needed. We will introduce the audience to Kripke
structures and to history based semantics (the latter due to RP and R. Ra
manujam). This semantics allows us to understand how knowledge grows as
a result of events and communication. The initial intended application wa
s to distributed computing\, but the tool has more general use.\n\nWe will
explain Jan Plaza's result on public announcements and some joint work wi
th Tasdemir and Witzel on how a knowledge manipulator can influence behavi
or by sending signals.\n\nIf time permits we will take a wider\, more phil
osophical perspective on knowledge\, explaining something about Plato's th
eory of knowledge as expounded in the Meno and the Theaetetus and the Gett
ier puzzles about knowledge.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=2654
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2654
END:VEVENT
BEGIN:VEVENT
SUMMARY:Epistemic Reasoning and Its Applications
DTSTART;VALUE=DATE-TIME:20130101T103000Z
DTEND;VALUE=DATE-TIME:20130101T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2655@cern.ch
DESCRIPTION:It is clear from examples from Shakespeare and others that kno
wledge and belief influence behavior.\n\nBut to understand these examples
some technical tools are needed. We will introduce the audience to\nKripk
e structures and to history based semantics (the latter due to RP and R. R
amanujam). This semantics\nallows us to understand how knowledge grows a
s a result of events and communication. The initial intended\napplication
was to distributed computing\, but the tool has more general use.\n\nWe w
ill explain Jan Plaza's result on public announcements and some joint work
with Tasdemir and Witzel on\nhow a knowledge manipulator can influence be
havior by sending signals.\n\nIf time permits we will take a wider\, more
philosophical perspective on knowledge\, explaining something about\nPlato
's theory of knowledge as expounded in the Meno and the Theaetetus and the
Gettier puzzles about\nknowledge.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=2655
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2655
END:VEVENT
BEGIN:VEVENT
SUMMARY:Program Analysis Using Quantifier Elimination Heuristics
DTSTART;VALUE=DATE-TIME:20130108T053000Z
DTEND;VALUE=DATE-TIME:20130108T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2694@cern.ch
DESCRIPTION:Loop invariants play a central role in ensuring the reliabilit
y of software. Program analysis techniques must either require such progra
m annotations at appropriate program locations or automatically derive the
se annotations from a program. A new approach for automatically generatin
g loop invariants from imperative programs will be presented. Loop invari
ants are assumed to have certain shape\, i.e.\, they are formulas in a res
tricted quantifier-free first-order theory. Elimination techniques can be
used to generate such invariants. The focus of this talk will be on explo
ring heuristics for quantifier-elimination so that such techniques can sca
le well. A nice feature of the proposed approach is that it does not need
to have access to any specification or annotations associated with a prog
ram. Some preliminary ideas about how to generalize these approaches to wo
rk on data types other than numbers will be discussed.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=2694
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2694
END:VEVENT
BEGIN:VEVENT
SUMMARY:Coordination over Networks: Simulation of Joint Distributions
DTSTART;VALUE=DATE-TIME:20130108T090000Z
DTEND;VALUE=DATE-TIME:20130108T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2698@cern.ch
DESCRIPTION:Networks are largely viewed as a medium for communication from
transmitters to receivers. More generally a network can be used as a medi
um to enable coordinated action between the entities represented by its no
des. Of particular interest is the simulation of joint distributions over
networks\, which is of interest both in security as in the creation of a s
hared secret key between some of the nodes of the network\, and in strateg
ic scenarios where the network\, as player in a game\, may need to jointly
randomize the actions of the entities that comprise it. There has been co
nsiderable recent interest and progress in the information theory communit
y on the problem of simulation of joint distributions over networks. We wi
ll discuss some of this progress\, present some of our new results\, and d
iscuss some of the open problems in this area (based on joint work with Vi
vek Borkar\, Amin Aminzadeh Gohari\, and Sudeep Kamath).\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=2698
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2698
END:VEVENT
BEGIN:VEVENT
SUMMARY:Spectrum Sensing for Cognitive Radios
DTSTART;VALUE=DATE-TIME:20130111T053000Z
DTEND;VALUE=DATE-TIME:20130111T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2662@cern.ch
DESCRIPTION:Inefficient use of radio spectrum is becoming a serious proble
m as more and more wireless systems are being developed to operate in crow
ded spectrum bands. Cognitive radio offers a novel solution to overcome th
e underutilization problem by allowing secondary usage of the spectrum res
ources along with high reliable communication. Spectrum sensing is a key e
nabler for cognitive radios. It identifies idle spectrum and provides awar
eness regarding the radio environment which are essential for the efficien
t secondary use of the spectrum and coexistence of different wireless syst
ems. The focus of this thesis is on the local and cooperative spectrum sen
sing algorithms. Local sensing algorithms are proposed for detecting ortho
gonal frequency division multiplexing (OFDM) based primary user (PU) trans
missions using their autocorrelation property. The proposed autocorrelatio
n detectors are simple and computationally efficient. Later\, the algorith
ms are extended to the case of cooperative sensing where multiple secondar
y users (SUs) collaborate to detect a PU transmission. For cooperation\, e
ach SU sends a local decision statistic such as log-likelihood ratio (LLR)
to the fusion center (FC) which makes a final decision. Cooperative sensi
ng algorithms are also proposed using sequential and censoring methods. Se
quential detection minimizes the average detection time while censoring sc
heme improves the energy efficiency. The performances of the proposed algo
rithms are studied through rigorous theoretical analyses and extensive sim
ulations. The distributions of the decision statistics at the SU and the t
est statistic at the FC are established conditioned on either hypothesis.
Later\, the effects of quantization and reporting channel errors are consi
dered. Main aim in studying the effects of quantization and channel errors
on the cooperative sensing is to provide a framework for the designers to
choose the operating values of the number of quantization bits and the ta
rget bit error probability (BEP) for the reporting channel such that the p
erformance loss caused by these non-idealities is negligible. Later a perf
ormance limitation in the form of BEP wall is established for the cooperat
ive sensing schemes in the presence of reporting channel errors. The BEP w
all phenomenon is important as it provides the feasible values for the rep
orting channel BEP used for designing communication schemes between the SU
s and the FC.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=2662
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2662
END:VEVENT
BEGIN:VEVENT
SUMMARY:Perturbations of Non-linear Systems of Differential Equations
DTSTART;VALUE=DATE-TIME:20130111T093000Z
DTEND;VALUE=DATE-TIME:20130111T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2714@cern.ch
DESCRIPTION:The method of variation of parameters\, credited to J.L. Lagra
nge\, expresses each solution of the perturbed system of differential equa
tions\, x' = A(t)x + f(t)\, in terms of the solution of the associated hom
ogeneous system\, y' = A(t)y. In this talk\, we will give an intuitive app
reciation of why this result is to be expected. Using this understanding\,
we will then discuss the analogue of this method\, developed by V.M. Alek
seev\, for obtaining solutions to perturbations of nonlinear system of dif
ferential equations.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=2714
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2714
END:VEVENT
BEGIN:VEVENT
SUMMARY:Uniqueness and Social Cost of Equilibria in Atomic Splittable Rout
ing Games
DTSTART;VALUE=DATE-TIME:20130201T090000Z
DTEND;VALUE=DATE-TIME:20130201T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2746@cern.ch
DESCRIPTION:In an atomic splittable routing game\, each player controls a
non-negligible\, splittable flow in a network. Each edge has a delay that
is a function of the total flow on the edge. Each player seeks a routing
strategy to minimize the total delay of his flow\, measured as the sum ove
r edges of his flow on the edge times that edge's delay. In this setting\,
a flow is at a Nash equilibrium if no player can unilaterally alter his
individual flow and reduce his total cost.\n\nIn this talk\, I will discus
s two topics about equilibria in atomic splittable routing games. The firs
t is the uniqueness of Nash equilibria. I will give a complete characteriz
ation on the multiplicity of equilibria based on graph topology. The secon
d topic is the social cost of equilibria when players form coalitions. In
particular\, I will talk about the conditions under which the post-collusi
on equilibrium are guaranteed to have less social cost than the pre-collus
ion equilibrium.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=2746
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2746
END:VEVENT
BEGIN:VEVENT
SUMMARY:One-shot Rate Distortion Under Maximum Distortion Criterion
DTSTART;VALUE=DATE-TIME:20130208T090000Z
DTEND;VALUE=DATE-TIME:20130208T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2765@cern.ch
DESCRIPTION:In information theory rate distortion addresses the problem of
determining the minimal number of bits per symbol\, as measured by the ra
te R\, that should be communicated over a channel\, so that the source (in
put signal) can be approximately reconstructed at the receiver (output sig
nal) without exceeding a given distortion D. In this talk we will discuss
rate distortion from the view point of one-shot (non-asymptotic) informati
on theory.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=2765
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2765
END:VEVENT
BEGIN:VEVENT
SUMMARY:Web Usage Mining and Recommender Systems
DTSTART;VALUE=DATE-TIME:20130215T060000Z
DTEND;VALUE=DATE-TIME:20130215T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2769@cern.ch
DESCRIPTION:Web Usage Mining is the application of data mining techniques
for discovery and analysis of user access patterns from the interaction ge
nerated by the users in the form of access logs\, proxy-server logs\, brow
ser logs\, etc. One major application is the use of these patterns in pred
iction and recommender systems to help personalize user browsing tasks. We
b usage mining consists of three phases\, namely preprocessing\, pattern d
iscovery and pattern analysis. This talk will first provide an overview of
the web usage mining and recommendation processes\, and then briefly desc
ribe some of the issues addressed in the research carried out with my stud
ents at Concordia.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=2769
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2769
END:VEVENT
BEGIN:VEVENT
SUMMARY:Cutting Graphs Using Eigenvectors\, a.k.a. Cheeger's Inequality
DTSTART;VALUE=DATE-TIME:20130215T090000Z
DTEND;VALUE=DATE-TIME:20130215T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2776@cern.ch
DESCRIPTION:How does one partition the vertex set of a graph into two part
s $(S\,S^C)$\,\nso that the ratio of edges going across to the Volume (num
ber of edges\nincident on vertices in S) in the vertex set is as small as
possible? This\nquantity is often referred to as the Sparsity $\\phi(S)$
of the cut\, and\nfinding the sparsest cut $\\phi(G)$ in a graph G is an i
mportant problem\,\nused as a subroutine in many algorithmic tasks\, like
spectral clustering.\nA widely-used method is based on the spectrum of the
Laplacian of the\ngraph\, and is referred to as spectral partitioning.\n\
nThe method is based on the following inequality: $ \\lambda_2 \\leq \\ph
i(G)\n\\leq O(\\sqrt{\\lambda_2}) $.\nHere $lambda_2$ is the second large
st eigenvalue of the normalized\nLaplacian of G.\n\nThis is a special case
of a more general inequality valid on Reimannian\nmanifolds proven by Che
eger\, and was adapted to the discrete setting by\nAlon and Milman.\n\nWe
shall see a proof of this inequality in the talk\, and how it can be\nused
to algorithmically generate a cut with a guarantee on sparsity as\nabove.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2776
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2776
END:VEVENT
BEGIN:VEVENT
SUMMARY:Bregman's Theorem
DTSTART;VALUE=DATE-TIME:20130222T090000Z
DTEND;VALUE=DATE-TIME:20130222T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2801@cern.ch
DESCRIPTION:Abstract: Let A be an $n \\times n$ boolean matrix\, i.e.\, it
s entries come\nfrom the set $\\{0\,1\\}$. Let $r_i$ denote the number of
1's in the row $i$.\nAlso\, let $S$ denote the set of permutations\, $\\si
gma$\, on $[n]$ such that\n$A_{i\, \\sigma(i)}=1$for all $i \\in [n]$. The
n the $perm(A)=|S|$. Bregman's\ntheorem is stated as follows:\n\n$$perm(A)
\\leq \\prod_{i\\in [n]}(r_i !)^{1/r_i}$$\n\nThis upper bounds the number
of perfect matchings in a bi-partite graph.\nWe are going to prove it.\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2801
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2801
END:VEVENT
BEGIN:VEVENT
SUMMARY:Market Games
DTSTART;VALUE=DATE-TIME:20130226T090000Z
DTEND;VALUE=DATE-TIME:20130226T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2778@cern.ch
DESCRIPTION:Markets are of course\, one of the oldest mechanism of exchang
ing commodities and cultural artifacts. While their history is over 5000 y
ears old\, their mathematical theory is rather recent. The first such mode
ls were by Fisher and Walras. Later models have been by Arrow\, Debreu and
others. Markets are now the central philosophical cornerstone of much of
the discourse on efficiency and development\, and it is imperative that w
e should investigate their functioning.\n\nWe begin by Fisher markets\, a
simple model of buyers with money and utilities\, and sellers with goods.
The standard efficient-market hypothesis predicts that unique prices are d
iscovered and the market clears. We study this market under strategic beha
viour by buyers who may report fictitious utlities. We show\, in fact\, th
at buyers may actually benefit from such behaviour. We then formulate the
Fisher market game and examine its Nash equilibria and show some not-so-su
rprising features.\n\nThe key to a market is its implementation\, whence w
e explore some standard laws (e.g.\, price discovery and the supply/demand
curve intersection) and their strategic basis. We close by highlighting t
wo commonly occuring market-games which need further research--the cowherd
of gokul game\, and the engineering college placement game.\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=2778
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2778
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Strahler Analysis of Binary Trees in Computer Science and in O
ther Sciences
DTSTART;VALUE=DATE-TIME:20130227T103000Z
DTEND;VALUE=DATE-TIME:20130227T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2802@cern.ch
DESCRIPTION:Computer scientists defined the Strahler number of a binary tr
ee in relation with the minimum number of registers needed for the comput
ation of an arithmetical expression. Beautiful asymptotic analysis for the
average Strahler number have been done (Flajolet\, Vuillemin\, Raoult\, K
emp)\, involving a periodic function coming from number theory. Surprising
ly\, this parameter appear in hydrogeology (Horton\, Strahler) for the mor
phologic study of river networks\, and also in molecular biology in the st
udy of RNA secondary structure(Waterman).\n\nApplications have been made i
n computer graphics (synthetic images of trees)\, the study of some fracta
l structures in experimental physics\, in radiology and in the domain of v
isualization of informations. Underlying this asymptotic analysis\, there
are deep combinatorial mathematics\, and some new structures have been int
roduced\, in collaboration with D.Knuth\, called Kepler towers. These obje
cts belongs to the "heaps of pieces" theory\, which gives a geometric inte
rpretation of equivalence classes of words in the so-called "trace monoid"
introduced in computer science by Mazurkiewicz as a model for concurrency
access to data structures.\n\nhttps://indico.tifr.res.in/indico/conferenc
eDisplay.py?confId=2802
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2802
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximation Algorithms for the Partition Vertex Cover Problem
DTSTART;VALUE=DATE-TIME:20130301T090000Z
DTEND;VALUE=DATE-TIME:20130301T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2815@cern.ch
DESCRIPTION:Let us consider a natural generalization of the Partial Vertex
Cover problem. Here\, an instance consists of a graph G=(V\,E)\, a cost f
unction c:V−>Z+\, a partition P1\,...\,Pr of the edge set E\, and a para
meter ki for each partition Pi. The goal is to find a minimum cost set of
vertices which cover at least ki edges from the partition Pi. We call this
the Partition-VC problem. In this talk\, you will see matching upper and
lower bounds on the approximability of this problem. The approximation alg
orithm that will be shown is based on a novel LP relaxation for this probl
em\, which is obtained by adding knapsack cover inequalities to a natural
LP relaxation of the problem. Further\, it will be shown that this LP has
an integrality gap of O(log r)\, where r is the number of sets in the part
ition of the edge set.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=2815
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2815
END:VEVENT
BEGIN:VEVENT
SUMMARY:Regenerations in Multiclass Open Queueing Networks
DTSTART;VALUE=DATE-TIME:20130305T060000Z
DTEND;VALUE=DATE-TIME:20130305T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2816@cern.ch
DESCRIPTION:Multiclass open queueing networks find wide applications in co
mmunication\, computer and fabrication networks. Often one is interested i
n steady state performance measures associated with these systems. One can
use the regenerative structure of the process to estimate the steady stat
e performance measures via regenerative simulation. In a queueing network\
, if all the interarrival times are Markovian (have exponential distributi
ons)\, it is easy to identify regeneration instants\, e.g.\, an instant co
rresponding to an arrival to an empty network. In this presentation\, we c
onsider networks where interarrival times are generally distributed but ha
ve exponential or fatter tails. We show that such distributions can be dec
omposed into mixture of sums of independent random variables such that at
least one of the components is exponentially distributed. This allows an e
mbedded regenerative structure in the Markov process. We show that under m
ild conditions on the network primitives\, the regenerative mean and stand
ard deviation estimators are consistent and satisfy a joint central limit
theorem. This is important as it allows construction of asymptotically val
id confidence intervals. We show that amongst all such interarrival time d
ecompositions\, the one with largest mean exponential component minimizes
the asymptotic variance of the standard deviation estimator.\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=2816
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2816
END:VEVENT
BEGIN:VEVENT
SUMMARY:Random Projections\, Margins and Learning
DTSTART;VALUE=DATE-TIME:20130308T090000Z
DTEND;VALUE=DATE-TIME:20130308T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2827@cern.ch
DESCRIPTION:In this talk we will give an introduction to random projection
s. Define the linear separability of data by a margin $\\gamma$. Then we w
ill see a simple algorithm to learn a linear separator and show that it wi
ll have error at most $\\frac{1}{2} - \\frac{\\gamma}{4}$ with probability
$\\Omega(\\gamma)$.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=2827
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2827
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Virtue of Succinct Proofs: Amplifying Communication Complex
ity Hardness to Time-Space Trade-offs in Proof Complexity
DTSTART;VALUE=DATE-TIME:20130313T100000Z
DTEND;VALUE=DATE-TIME:20130313T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2823@cern.ch
DESCRIPTION:An active line of research in proof complexity over the last d
ecade has been the study of proof space and trade-offs between size and sp
ace. Such questions were originally motivated by practical SAT solving\, b
ut have also led to the development of new theoretical concepts in proof c
omplexity of intrinsic interest and to results establishing nontrivial rel
ations between space and other proof complexity measures.\n\nBy now\, the
resolution proof system is fairly well understood in this regard\, as witn
essed by a long sequence of papers leading up to [Ben-Sasson and Nordstrom
2008\, 2011] and [Beame\, Beck\, and Impagliazzo 2012]. However\, for oth
er relevant proof systems in the context of SAT solving\, such as polynomi
al calculus (PC) and cutting planes (CP)\, very little has been known.\n\n
Inspired by [BN08\,BN11]\, we consider CNF encodings of so-called pebble g
ames played on graphs and the approach of making such pebbling formulas ha
rder by simple syntactic modifications. We use this paradigm of hardness a
mplification to make progress on the relatively longstanding open question
of proving time-space trade-offs for cutting planes and polynomial calcul
us. Specifically\, we exhibit a family of formulas F_nsuch that:\n\n- F_n
has size Theta(n) and width O(1).\n- F_n has a resolution refutation in le
ngth O(n) which generalizes to both CP and PC.\n- For any refutation of F_
n in CP or PCR (an extension of PC) in length L and space s it holds that
s log L is at least (roughly) sqrt[4]{n}.\n\nA crucial technical ingredien
t in our result is a new two-player communication complexity lower bound f
or composed search problems in terms\nof block sensitivity\, a contributio
n that we believe to be of independent interest.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=2823
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2823
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Manipulablilty of Elections on Average
DTSTART;VALUE=DATE-TIME:20130315T090000Z
DTEND;VALUE=DATE-TIME:20130315T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2839@cern.ch
DESCRIPTION:Elections are a way of aggregating preferences made by individ
ual voters and making a choice for the whole. Questions about how to make
such a "social choice"\, or whether such a choice can be made have been we
ll studied under the topic of social choice theory. One of the staggering
conclusions(known as the Arrows Imposibility theorem) derived very early
was that\, if we assume certain simple assumptions about elections\, then
the only way of making the social choice is a dictatorship. In this lectur
e\, we will see a quantitative version of a similar theorem(Gibbard–Satt
erthwaite theorem). That is\, for a random voting profile\, with non trivi
al probability an assumption about manipulability of the election is viola
ted if the election is very different from a dictatorship.\n\nFor the supe
r interested\, we will be covering the paper "The Geometry of Manipulation
- a Quantitative Proof of the Gibbard Satterthwaite Theorem" by Isaksson\
, Mossel and Kindler. The proof is purely geometric\, combinatorial and do
es not use discrete harmonic analysis.\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=2839
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2839
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stochastic Strategies in Minority Game
DTSTART;VALUE=DATE-TIME:20130322T090000Z
DTEND;VALUE=DATE-TIME:20130322T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2859@cern.ch
DESCRIPTION:Minority Game was introduced as a model for competition betwe
en interacting agents in scarce resource conditions. In the game\, an odd
number of players each must choose one of two possible choices independen
tly at each turn and the players who end up on the minority side win. I wi
ll discuss several features of this game with special emphasis on the cas
e where players use stochastic strategies.\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=2859
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2859
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computation and Secrecy in Multiterminal Systems
DTSTART;VALUE=DATE-TIME:20130402T053000Z
DTEND;VALUE=DATE-TIME:20130402T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2840@cern.ch
DESCRIPTION:nformation theoretic secrecy provides a framework for explorin
g schemes that guarantee provable unconditional security in network system
s. This talk explores innate structural connections that exist between the
information theoretic notion of multiterminal secrecy on the one hand and
data compression and function computation on the other. For a setup where
multiple terminals observe separate but correlated data and communicate o
ver a public channel\, three classes of secrecy problems are studied: (i)
generating secret keys\, (ii) ensuring secrecy against a querying eavesdro
pper\, and (iii) secure function computation over an insecure communicatio
n network. The aforementioned connections lead to new results and algorith
ms for all of these problems. Applications abound\, including in biometric
security\, hardware security\, cloud computing and in-network computation
in sensor networks. This talk is based on different joint works with Prak
ash Narayan\, Piyush Gupta\, Navin Kashyap\, Yogesh Sankarasubramaniam and
Kapali Viswanathan.\n\nBiography: Himanshu Tyagi received his Bachelor of
Technology degree in Electrical Engineering and Master of Technology degr
ee in Communications and Information Technology\, both from the Indian Ins
titute of Technology\, Delhi in 2007. His master 's thesis advisor was Pro
f. Ranjan K. Mallik. He is currently a doctoral candidate in the Departmen
t of Electrical and Computer Engineering at the University of Maryland\, C
ollege Park. His Ph.D. advisor is Prof. Prakash Narayan. He has interned a
t National Semiconductors\, Bangalore in 2005\, at Telecommunications Rese
arch Center Vienna\, Vienna in 2006 (with Dr. Jossy Sayir)\, at Bell Labs\
, New Jersey in 2009 (with Dr. Piyush Gupta)\, and has spent a semester at
Alfrèd Rènyi Institute of Mathematics in 2010 (with Prof. Imre Csiszár
). He is a Clark School's Future Faculty Fellow (2010)\, and an ECE Distin
guished Dissertation Fellow (2012) at the University of Maryland\, College
Park. Also\, he was a finalist for the best student paper award at the I
EEE International Symposium on Information Theory (2010).\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=2840
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2840
END:VEVENT
BEGIN:VEVENT
SUMMARY:Newman's Theorem
DTSTART;VALUE=DATE-TIME:20130405T090000Z
DTEND;VALUE=DATE-TIME:20130405T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2889@cern.ch
DESCRIPTION:In the literature of communication complexity\, two models of
random protocols are used - common random string model and private random
string model. In a common random string model\, Alice and Bob have access
to a common random string which they use in their calculation. On the cont
rary\, in private random string model\, Alice or Bob cannot have access to
other person's private random string.\n\nIt is trivial to see that common
random string model subsumes private random string model as any private c
oin protocol can be simulated by a public coin protocol where the common r
andom string is nothing but concatenation of the private random strings of
Alice and Bob. The interest question to ask is whether the opposite direc
tion is possible.\n\nWe investigate the relative power of these models and
show that the two models are essentially equal.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=2889
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2889
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computing the Shape of Point Clouds
DTSTART;VALUE=DATE-TIME:20130410T103000Z
DTEND;VALUE=DATE-TIME:20130410T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2836@cern.ch
DESCRIPTION:A point cloud is an unorganized collection of a very large num
ber of points usually with just their coordinates and no other information
. Such point clouds are increasingly encountered in the form of outputs fr
om 3D scanning devices\, 3D from images/sketches\, from physically-based s
imulations\, etc. Computing the shape of a point cloud (a curve in 2D and
surface in 3D)\, is known to be a difficult problem. Imagine the game of c
onnecting the dots but with no labels on the points and extend it to highe
r dimensions. This problem has been the subject of a lot of research over
the last 3 decades with many different algorithmic solutions proposed. A r
obust solution has many applications such as reverse engineering\, 3D prin
ting\, visualization\, machine learning\, etc. When posed as a search for
the “desired” shape\, this problem is most probably NP-hard. Apart fro
m efficiency\, the other main challenges include ability to handle non-den
se and non-uniform sampling\, accuracy to sharp features\, robustness and
provable reconstruction guarantees – that is\, what is the class of poin
t clouds for which the algorithm yields correct results? In this talk\, w
e will first introduce the problem and its challenges\, then describe some
of the better known previously proposed solutions and conclude with a bri
ef presentation of work done with my student.\n \n\nBrief Bio:\n\nDr. S. P
. Mudur is presently visiting the School of Technology and Computer Scienc
e at TIFR. He got his B.Tech. from IIT Bombay in 1970 and his PhD from TIF
R (through University of Mumbai) in 1976. He was with TIFR from 1970 – 1
985 and with the National Centre for Software Technology (NCST) from 1985-
2002. Since 2002 he is a professor in the computer science and software e
ngineering department of Concordia University in Montreal\, Canada. Dr. Mu
dur started his research in computer graphics at TIFR in the early 70’s
and has continued to do research in that field. His primary focus is on pr
oblems in 3D modeling and animation.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=2836
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2836
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Relation Between Stochastic and Deterministic Models for Chemi
cal Reactions
DTSTART;VALUE=DATE-TIME:20130412T090000Z
DTEND;VALUE=DATE-TIME:20130412T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2908@cern.ch
DESCRIPTION:+ Tom Kurtz\, in a classic paper from 1972\, showed that stoch
astic models for chemical reactions and deterministic models are related:
in the limit of volume going to infinity\, the stochastic model approaches
the deterministic model. \n\n+ More recently\, in 2010\, Anderson\, Cra
ciun\, and Kurtz showed that stochastic models for "complex-balanced" chem
ical reaction systems admit a product-form stationary distribution. This g
eneralizes a result in queueing theory due to Kelly in 1979.\n\nI will des
cribe these two results.\n\nREFERENCE : Continuous Time Markov Chain Model
s for Chemical Reaction Networks by David F. Anderson and Thomas G. Kurtz.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2908
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2908
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Model for Equilibrium in Some Service-provider User-set Interact
ions
DTSTART;VALUE=DATE-TIME:20130416T093000Z
DTEND;VALUE=DATE-TIME:20130416T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2913@cern.ch
DESCRIPTION:We propose a model for interaction between an user-set (market
) and a service-provider (firm) when the offered demand is sensitive to th
e offered Quality of Service. We show that\, under fairly mild conditions\
, an equilibrium point or an equilibrium set exists when the service-provi
der bases its operational decisions as per an optimal policy of an average
reward multi-chain Markov decision model. We illustrate these and related
issues via a model for vehicle relocation system.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=2913
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2913
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Survey on Uniform Hardness Amplification in NP
DTSTART;VALUE=DATE-TIME:20130417T040000Z
DTEND;VALUE=DATE-TIME:20130417T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2920@cern.ch
DESCRIPTION:We investigate the following question: If NP contains function
s which are slightly hard to compute on average then does NP also contain
function which is very hard to compute on average? Similar result is shown
to hold outside NP by Yao by his celebrated XOR lemma. But same technique
breaks down to amplify the hardness inside NP.\n\nWe will restrict our at
tention to boolean function though the result that we present can be gener
alized to functions that are not boolean valued. O'Donnell [OD02] showed h
ardness can be amplified in NP in non-uniform setting and he amplified the
hardness from $1/poly(n)$ to $(1/2-1/n^{1/2+\\epsilon})$ for polynomial s
ized circuit. Healy et al [HVV04] extended this result and amplified the h
ardness to the optima.\n\nTrevisan in his papers [Tre03\, Tre05] amplified
the hardness from $(1/poly(n))$ to $(1/2 - 1/poly \\log (n))$ in uniform
setting. The results of Trevisan can be explained in terms of error correc
ting codes as done by Buresh-Oppenheim et al [BKS06]. This\, in turn\, lea
ds to the open question of finding a monotone code with efficient list dec
oding algorithm.\n\nIn this talk\, we will describe the gradual developmen
t of hardness amplification in NP.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=2920
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2920
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fast Low Rank Approximation via Random Projections
DTSTART;VALUE=DATE-TIME:20130419T090000Z
DTEND;VALUE=DATE-TIME:20130419T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2929@cern.ch
DESCRIPTION:Given an m x n matrix A\, we define the rank-k approximation o
f A as a matrix B of same size and of rank at most k such that A and B are
close in the Frobenius norm. In this talk\, we will first give a geometri
c interpretation of this problem and relate it to the singular value of de
composition (SVD) of the matrix A. We will then use this knowledge to unde
rstand why a fast low rank approximation algorithm is to be expected. If t
ime permits\, we will see some applications of low rank approximations to
a class of fixed point and minimization problems.\n\nhttps://indico.tifr.r
es.in/indico/conferenceDisplay.py?confId=2929
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2929
END:VEVENT
BEGIN:VEVENT
SUMMARY:Bridging the Gap: From Non-determinism to Determinism in UL
DTSTART;VALUE=DATE-TIME:20130426T090000Z
DTEND;VALUE=DATE-TIME:20130426T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2942@cern.ch
DESCRIPTION:The class of Unambiguous Star-free regular languages over word
s has been variously characterized in the past by logics such as the fragm
ents \\Delta_2[\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=2942
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2942
END:VEVENT
BEGIN:VEVENT
SUMMARY:Local Search for Computationally Hard Constraint Satisfaction Prob
lems
DTSTART;VALUE=DATE-TIME:20130429T050000Z
DTEND;VALUE=DATE-TIME:20130429T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2941@cern.ch
DESCRIPTION:Since the emergence of Artificial Intelligence (AI) as a new f
ield of modern science and engineering about 6 decades ago\, concerted ef
forts have been made on designing and developing expressively adequate lan
guages to represent knowledge about real world domains\, and building comp
utational tools to efficiently reason with these representations. Indeed\,
these two aspects of intelligent systems remain the fundamental challenge
s of AI. This talk will first give an overview of the AI challenge in gen
eral. We will then focus on the constraint satisfaction paradigm that has
become a powerful approach to model real world problems and solve them eff
iciently using general purpose constraint solving techniques. We will pres
ent some of our recent successes on solving the propositional satisfiabili
ty problems\, and some open issues.\n\nBio: Professor Abdul Sattar is the
founding Director of the Institute for Integrated and Intelligent Systems
and a Professor of Computer Science and Artificial Intelligence at Griffi
th University. He is also a Research Leader at National ICT Australia (NIC
TA) Queensland Research Lab (QRL) where he held the position of QRL Edu
cation Director\, the leader of the SAFE Agents project (July 2005-June 20
08)\, and currently leading the QRL node of the NICTA's largest project\,
Advance Technologies for Optimisation and Modelling In Constraints (ATOMIC
). He has been an academic staff member at Griffith University since Febr
uary 1992 as a lecturer (1992-95)\, senior lecturer (1996-99)\, and profes
sor (2000-present) within the School of Information and Communication Tech
nology. Prior to his career at Griffith University\, he was a lecturer in
Physics in Rajasthan\, India (1980-82)\, research scholar at Jawaharlal N
ehru University\, India (1982-85)\, the University of Waterloo\, Canada (1
985-87)\, and the University of Alberta\, Canada (1987-1991).\n\nHe has wo
n several awards starting from the national scholarships during his studie
s in India (1974-79)\, a Commonwealth scholarship in Canada (1985-1990)\,
to a number of research grants. The main grants include 7 ARC Large/Discov
ery grants (worth approximately $1.5M). He has published over 120 refereed
papers in international journals and conferences\, a large number of thes
e papers appeared in A* and A category conferences and journals. His resea
rch team has won 3 international awards in recent years including IJCAI 20
07 Distinguished Paper award\, Gold Medals in the 2005 and 2007 SAT solver
competitions. He has supervised successful completion of 13 PhD students
as a principal supervisor.\n\nHis current research interests include knowl
edge representation and reasoning\, constraint satisfaction\, intelligent
scheduling\, rational agents\, propositional satisfiability\, temporal rea
soning\, temporal databases\, and computational proteomics.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=2941
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2941
END:VEVENT
BEGIN:VEVENT
SUMMARY:Boolean Monotonicity Testing via an Isoperimetry Result on the Dir
ected Hypercube
DTSTART;VALUE=DATE-TIME:20130502T100000Z
DTEND;VALUE=DATE-TIME:20130502T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2952@cern.ch
DESCRIPTION:A Boolean function f:{0\,1}^n \\to {0\,1} is monotone if f(x)
\\geq f(y) whenever x > y\, that is\, all coordinates of x dominate those
of y. f is \\eps-far from monotone if the function needs to be changed in
at least \\eps fraction of the domain points to make it monotone. We are i
nterested in the problem of distinguishing monotone functions from those t
hat are \\eps-far\, given only query access to the function\, making as fe
w queries to the function as possible. There is a relatively easy tester m
aking O(n/\\eps)-queries from 2000. In this talk\, I'll show a slightly mo
re sophisticated tester which makes o(n) queries (to be precise\, it makes
O(n^{5/6}/\\eps^{10/3}) queries).\n\nOne of the main ingredients will be
an isoperimetry result on the *directed* hypercube\, where the edges of th
e hypercube are directed from y to x if x > y. In particular\, we show tha
t the support set (the x's s.t. f(x)=1) of a \\eps-far from monotone func
tion\, either have a large directed edge expansion\, or a large directed v
ertex expansion (joint work with C. Seshadhri).\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=2952
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2952
END:VEVENT
BEGIN:VEVENT
SUMMARY:Ramanujan Graphs of All Degrees
DTSTART;VALUE=DATE-TIME:20130506T110000Z
DTEND;VALUE=DATE-TIME:20130506T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2921@cern.ch
DESCRIPTION:We prove that there exist infinite families of bipartite Raman
ujan graphs of every degree bigger than 2. We do this by proving a variant
of a conjecture of Bilu and Linial about the existence of good 2-lifts of
every graph. We also construct infinite families of `irregular Ramanujan'
graphs\, whose eigenvalues are bounded by the spectral radius of their un
iversal cover. Such families were conjectured to exist by Linial and other
s. In particular\, we construct infinite families of (c\,d)-biregular bipa
rtite graphs with all non-trivial eigenvalues bounded by $\\sqrt{c-1}+\\sq
rt{d-1}$\, for all $c\, d \\geq 3$. Our proof exploits a new technique for
demonstrating the existence of useful combinatorial objects that we call
the "Method of Interlacing Polynomials" (joint work with A. Marcus and D.
Spielman).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=2921
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2921
END:VEVENT
BEGIN:VEVENT
SUMMARY:Automated Analysis of Probabilistic Infinite-state Systems
DTSTART;VALUE=DATE-TIME:20130510T090000Z
DTEND;VALUE=DATE-TIME:20130510T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2981@cern.ch
DESCRIPTION:I will survey the state of the art in the automated analysis o
f several classes of probabilistic infinite-state systems which naturally
arise when studying probabilistic programs with recursive calls\, queueing
systems\, population dynamics and natural language processing. I will des
cribe the relationships between these different models and recent developm
ents in the computational complexity of their analysis. I will also briefl
y describe tool PReMo\, which is able to analyse these models\, along with
some interesting performance results. (This talk is based on several join
t papers with T. Brazdil\, V. Brozek\, K. Etessami\, S. Kiefer\, A. Kucera
\, and M. Yannakakis).\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=2981
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2981
END:VEVENT
BEGIN:VEVENT
SUMMARY:Topology of Some Random Complexes
DTSTART;VALUE=DATE-TIME:20130514T053000Z
DTEND;VALUE=DATE-TIME:20130514T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2987@cern.ch
DESCRIPTION:The field of topological data analysis has received attention
in the recent years and a key object of study are random simplicial comple
xes. After a brief introduction to the field\, we shall look at the more s
pecific case of simplicial complexes built over random points in space. We
shall in particular try to see how dependence structure of the random poi
nts affect the topology of the random simplicial complexes. We shall also
try to illustrate the proof techniques rely on the study of the correspond
ing the more well-known random geometric graphs. Only a basic knowledge of
probability and topology will be assumed (this is a joint work with Prof.
Robert Adler).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=2987
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2987
END:VEVENT
BEGIN:VEVENT
SUMMARY:Probabilistic Approximation of Metrics by Tree Metrics
DTSTART;VALUE=DATE-TIME:20130517T090000Z
DTEND;VALUE=DATE-TIME:20130517T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2993@cern.ch
DESCRIPTION:Consider a finite set $V$ and a function $d: V \\times V \\rig
htarrow \\mathbb{R}$ such that $(V\,d)$ forms a metric space. A function $
d_T : V' \\times V' \\rightarrow \\mathbb{R}$ is called a tree metric on $
V'$ if for all $u\,v \\in V'$\, $d_T(u\,v)$ is the length of the unique sh
ortest path between them in $T$\, where $T$ is a tree on $V'$ with nonnega
tive lengths on its edges. A well studied problem in literature is to appr
oximate a given metric $d$ on $V$ using some tree metric $d_T$ on $V' \\su
pseteq V$ in the sense that $d(u\,v) \\le d_T(u\,v) \\le \\alpha d(u\,v)$
for all $u\,v \\in V$ and for some $\\alpha$. In this talk\, I will presen
t the result of Fakcharoenphol et al. that there is a randomized polynomia
l time algorithm which\, given any metric $d$ on an $n$-sized set $V$\, pr
oduces a tree metric $d_T$ on some $V' \\supseteq V$\, such that for all $
u\,v \\in V$\, $d(u\,v) \\le d_T(u\,v)$ and $E[d_T(u\,v)] \\le O(\\log n)
d(u\,v)$.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId
=2993
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2993
END:VEVENT
BEGIN:VEVENT
SUMMARY:Probabilistic Programming: A Program Analysis Perspective
DTSTART;VALUE=DATE-TIME:20130521T060000Z
DTEND;VALUE=DATE-TIME:20130521T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2990@cern.ch
DESCRIPTION:Probabilistic models\, particularly those with causal dependen
cies\, can be succinctly written as probabilistic programs. Recent years h
ave seen a proliferation of languages for writing such probabilistic progr
ams\, as well as tools and techniques for performing inference over these
programs In this talk\, we show that static and dynamic program analysis t
echniques can be used to infer quantities of interest to the machine learn
ing community\, thereby providing a new and interesting domain of applicat
ion for program analysis. In particular\, we show that static analysis tec
hniques inspired by dataflow analysis and iterative refinement techniques
can be used for Bayesian inference. We also show that dynamic analysis tec
hniques inspired by directed testing and symbolic execution can be used fo
r sampling probabilistic programs (joint work with Aditya Nori\, Microsoft
Research India).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.p
y?confId=2990
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2990
END:VEVENT
BEGIN:VEVENT
SUMMARY:Synergistic Challenges in Data-Intensive Science and Exascale Comp
uting
DTSTART;VALUE=DATE-TIME:20130522T053000Z
DTEND;VALUE=DATE-TIME:20130522T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-2999@cern.ch
DESCRIPTION:The two dominant paradigms for scientific discovery have histo
rically been theory and experiments\, with large-scale simulations emergin
g as the third paradigm in the 20th century. Over the past decade\, a new
paradigm for scientific discovery is emerging due to the availability of e
xponentially increasing volumes of data from large instruments such as tel
escopes\, colliders\, and light sources\, as well as the proliferation of
sensors and high-throughput analysis devices. These trends are popularly
referred to as "Big Data"\, and they have led to the emergence of a "fourt
h paradigm" for scientific discovery that is driven by analysis of massive
datasets to extract new knowledge and actionable insights. The complexity
and challenge of the fourth paradigm arises from the increasing velocity\
, heterogeneity\, and volume of data generation.\n \nAnalysis of large vol
umes of complex data to derive knowledge requires "data-driven" computing\
, where the computation is driven by data mining\, queries\, analysis\, st
atistics\, and hypothesis formulation and validation. At the same time\,
large scale simulations that are expected to run on future exascale ("Big
Compute") systems are accompanied by the challenges of data-intensive comp
uting. The requirements for Big Data and Big Compute are tightly intertwi
ned since they both contribute to a shared goal of scientific discovery. F
or example\, data-intensive simulations on Big Compute exascale systems wi
ll be used to generate volumes of Big Data that are comparable to the data
volumes generated by many scientific instruments. Likewise\, the volumes
of Big Data generated by the data-driven paradigm will need to be analyze
d by Big Compute exascale or extreme-scale systems. As we head to the exa
scale timeframe\, it will be critical to exploit synergies between the two
approaches\, even though some fundamental differences may still remain be
tween the two.\n \nThis talk will discuss some of these synergies in the c
ontext of challenges in data-intensive science and exascale computing. Th
e material for this talk is drawn from a recent (March 2013) study led by
the speaker on "Synergistic Challenges in Data-Intensive Science and Exasc
ale Computing" for the US Department of Energy's Office of Science. Backg
round material was also drawn from an earlier (September 2009) DARPA Exasc
ale Software Study\, also led by the speaker. We would like to acknowledg
e the contributions of all participants in both studies.\n \nBIO: Vivek Sa
rkar conducts research in multiple aspects of parallel software including
programming languages\, program analysis\, compiler optimizations and runt
imes for parallel and high performance computer systems. He currently lea
ds the Habanero Multicore Software Research project at Rice University\, a
nd serves as Associate Director of the NSF Expeditions project on the Cent
er for Domain-Specific Computing. Prior to joining Rice in July 2007\, Viv
ek was Senior Manager of Programming Technologies at IBM Research. His re
sponsibilities at IBM included leading IBM's research efforts in programmi
ng model\, tools\, and productivity in the PERCS project during 2002- 2007
as part of the DARPA High Productivity Computing System program. His pas
t projects include the X10 programming language\, the Jikes Research Virtu
al Machine for the Java language\, the MIT RAW multicore project\, the AST
I optimizer used in IBM's XL Fortran product compilers\, the PTRAN automat
ic parallelization system\, and profile-directed partitioning and scheduli
ng of Sisal programs. Vivek holds a B.Tech. degree from the Indian Instit
ute of Technology\, Kanpur\, an M.S. degree from University of Wisconsin-M
adison\, and a Ph.D. from Stanford University. He became a member of the I
BM Academy of Technology in 1995\, the E.D. Butcher Chair in Engineering a
t Rice University in 2007\, and was inducted as an ACM Fellow in 2008. Vi
vek has been serving as a member of the US Department of Energy's Advanced
Scientific Computing Advisory Committee (ASCAC) since 2009.\n \nREFERENCE
S:\n \nhttp://science.energy.gov/~/media/ascr/ascac/pdf/reports/2013/ASCAC
_Data_Intensive_Computing_report_final.pdf\nhttp://www.cs.rice.edu/~vsarka
r\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2999
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=2999
END:VEVENT
BEGIN:VEVENT
SUMMARY:Correctness of Closed Nesting in Software Transactional Memory Sys
tems
DTSTART;VALUE=DATE-TIME:20130522T103000Z
DTEND;VALUE=DATE-TIME:20130522T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3003@cern.ch
DESCRIPTION:There has been considerable interest in Software Transactional
Memory in recent years. The reason for rise in STM is due to rise of mult
icore computers. To fully utilize the power of these machines\, applicatio
ns need to be able to harness the parallelism of the underlying hardware.
This is commonly achieved using multi-threading. Yet writing correct and s
calable multi-threaded programs is far from trivial. In multi-threaded pro
grams sets of semantically related actions may need to execute in mutual e
xclusion to avoid semantic inconsistencies.\n \nTraditionally\, multi-thre
aded programs were developed in conjunction with locks to address these is
sues. But programming with locks has many disadvantages such as deadlocks\
, priority inversion etc. and makes it difficult to build scalable softwar
e systems. Importantly\, lock based software components are difficult to c
ompose i.e. build larger software systems using simpler software component
s. Composition of software components is a very important property which i
s the basis of modular programming. Nesting of transactions is a way of ac
hieving Composition. \n \nIn this talk\, I will explain about how closed n
esting can be achieved with Software Transactional Memory. I will talk abo
ut correctness requirements of Closed Nested executions. I will then descr
ibe correctness criteria that we developed and then describe about an impl
ementation based on them.\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=3003
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3003
END:VEVENT
BEGIN:VEVENT
SUMMARY:How Easy/Hard it is to Schedule in Networks?
DTSTART;VALUE=DATE-TIME:20130524T090000Z
DTEND;VALUE=DATE-TIME:20130524T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3007@cern.ch
DESCRIPTION:In the context of scheduling in networks\, we have the famous
max-weight scheduling policy which minimizes delays\, but is computational
ly intensive for large networks. We also have policies that are computatio
nally nice but offer no guarantees on delays. The question now is: Can we
have a poly time computable scheduling policy that achieves "low" delays?
We shall answer this by considering two useful models of communication ne
tworks: the independent set constraints model and the SINR model.\n\nRefer
ence: Devavrat Shah\, David N. C. Tse\, and John N. Tsitsiklis. 2011. Hard
ness of Low Delay Network Scheduling. IEEE Trans. Inf. Theor. 57\, 12 (Dec
ember 2011)\, 7810-7817.\n\nhttps://indico.tifr.res.in/indico/conferenceDi
splay.py?confId=3007
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3007
END:VEVENT
BEGIN:VEVENT
SUMMARY:Navigating the Deletion Channel
DTSTART;VALUE=DATE-TIME:20130606T053000Z
DTEND;VALUE=DATE-TIME:20130606T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3015@cern.ch
DESCRIPTION:A deletion channel is communication medium that takes a string
of symbols as input\, deletes a fixed number of them and outputs the rema
ining symbols by aligning them without changing their order. Importantly\,
the position of the deleted symbol is not known at the output. A deletion
-correcting code is a set of input strings with pairwise disjoint output s
ets. The central object of interest to us is the largest deletion correcti
ng code for input strings of the same length. Despite many years of effort
on this problem it remains open at a fundamental level.\n\nA specific ver
sion of this problem\, however\, holds promise for a complete solution. Fo
r the case of binary strings and a single deletion\, a number-theoretic co
nstruction is known\, which partitions the 2^n binary strings of length n
into n+1 classes\, each of which is a single-deletion correcting code (cal
led the Varshamov-Tenengolts codes). The largest of these classes is known
to asymptotically optimal.\n\nThis seminar will discuss our recent effort
s at solving this problem. We show that if suitably viewed the Varshamov-T
enengolts codes are simultaneously responsible for the solution of several
combinatorial problems that appear to be otherwise unrelated. For the bin
ary single-deletion case we find a new code construction that is asymptoti
cally optimal. We obtain the first ever nonasymptotic upper bounds on size
s of the largest codes for arbitrary number of deletions and arbitrary alp
habet size.\n\nRemarkably\, neither of our these results (or their proofs)
show any number-theoretic character\, despite the fact that the most prom
ising code is number-theoretic. My aim in this seminar is to share these n
ew results\, speculate on possible connections and invite others to think
on this problem.\n\nBio: Ankur is an Assistant Professor with the Systems
and Control Engineering group at IIT Bombay and a recipient of the INSPIRE
Faculty Award of the Department of Science and Technology\, Government of
India\, 2013. He received his B.Tech. in Aerospace Engineering from IIT B
ombay in 2006\, M.S. in General Engineering in 2008 and Ph.D. in Industria
l Engineering in 2010\, both from the University of Illinois at Urbana-Cha
mpaign (UIUC). From 2010–2012 he was a post-doctoral researcher at the C
oordinated Science Laboratory at UIUC. His research interests include game
theory and economics\, optimization and variational inequalities\, combin
atorial coding theory problems\, the role of information in stochastic con
trol\, and operations research.\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=3015
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3015
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lower Bound Techniques in Property Testing
DTSTART;VALUE=DATE-TIME:20130607T090000Z
DTEND;VALUE=DATE-TIME:20130607T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3020@cern.ch
DESCRIPTION:As with other areas of computer science\, lower bounds play an
essential role in the field of property testing. The question can be pose
d as follows. Given a black-box access to an arbitrary function which is t
o be tested against a property\, how many queries are necessary to be made
to the function so that with high probability the function can be disting
uished between having the property and ϵ-away from all the functions havi
ng the property.\n\nThe tester can be both adaptive and non-adaptive. The
most efficient and generic tool for proving lower bounds for property test
ing was given by Yao. We will look at the statement and proof of these too
ls and their application in proving lower bounds of the query complexity o
f testing k-linearity and monotonicity.\n\nReferences:\n\n[BK12] Eric Bla
is and Daniel M. Kane. Tight bounds for testing k-linearity. In Anupam Gup
ta\, Klaus Jansen\, Jos´e D. P. Rolim\, and Rocco A. Servedio\, editors\,
APPROX-RANDOM\, volume 7408 of Lecture Notes in Computer Science\, pages
435–446. Springer\, 2012.\n\n[FLN+02] Eldar Fischer\, Eric Lehman\, Ilan
Newman\, Sofya Raskhodnikova\, Ronitt Rubinfeld\, and Alex Samorodnitsky.
Monotonicity testing over general poset domains. In John H. Reif\, editor
\, STOC\, pages 474–483. ACM\, 2002.\n\n[Yao77] Andrew Chi-Chih Yao. Pro
babilistic computations: Toward a unified measure of complexity (extended
abstract). In FOCS\, pages 222–227. IEEE Computer Society\, 1977.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3020
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3020
END:VEVENT
BEGIN:VEVENT
SUMMARY:50 Years of the Krohn-Rhodes Theorem
DTSTART;VALUE=DATE-TIME:20130612T053000Z
DTEND;VALUE=DATE-TIME:20130612T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3021@cern.ch
DESCRIPTION:A little over 50 years ago (1962)\, we had the first nontrivia
l theorem which used an algebraic approach to automata theory: Schuetzenbe
rger's theorem giving an algorithm to check whether a given regular langua
ge is definable using a starfree expression. The same year saw the announc
ement of the extension of the classification of finite groups to finite se
migroups: the Krohn-Rhodes theorem showing that the semigroups which are g
roupfree can be composed using just one three-element "prime"\, the remain
ing primes being the finite simple groups. The journal paper appeared in 1
965. (The complexity of this decomposition remains open.) In 1997\, a new
proof was found for Schuetzenberger's theorem. Recently this has led to ne
w proofs for the Krohn-Rhodes theorem. It would be too much to pack all th
is into one talk\, so we will give an overview\, and explain all the requi
red algebra.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=3021
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3021
END:VEVENT
BEGIN:VEVENT
SUMMARY:Parikh's Theorem
DTSTART;VALUE=DATE-TIME:20130614T090000Z
DTEND;VALUE=DATE-TIME:20130614T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3031@cern.ch
DESCRIPTION:We know that every regular language is context-free but a cont
ext-free language need not be regular. In 1966\, Rohit Parikh showed that
if th order of the symbols in a context-free language is ignored\, it is i
mpossible to distinguish it from a regular set\, thus implying that contex
t-free languages can have a richer structure than those obtained rom regul
ar sets.\n\nParikh used properties of semi-linear sets and derivation tree
s and presented some interesting combinatorial arguments to arrive at his
result. Although Goldstine came up with a simplified proof 11 years later\
, Parikh's theorem is still considered remarkable since the exact conditio
ns controlling the structure of the derivation trees that he defined are n
ontrivial.\n\nIn this talk\, we will define these terms and use them to pr
ove Parikh's theorem\, and time permitting\, discuss some of its applicati
ons.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3031
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3031
END:VEVENT
BEGIN:VEVENT
SUMMARY:Decision Trees and the ID3 Algorithm
DTSTART;VALUE=DATE-TIME:20130621T090000Z
DTEND;VALUE=DATE-TIME:20130621T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3043@cern.ch
DESCRIPTION:Decision trees are used in machine learning a lot. In this tal
k I will give a brief introduction to decision trees. Then we will look at
the ID3 (Iterative Dichotomiser 3) algorithm by Ross Quinlan used to gene
rate a decision tree from a dataset.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=3043
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3043
END:VEVENT
BEGIN:VEVENT
SUMMARY:Evolvability as Learnability
DTSTART;VALUE=DATE-TIME:20130626T090000Z
DTEND;VALUE=DATE-TIME:20130626T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3046@cern.ch
DESCRIPTION:150 years after Darwin's seminal theory\, there are many aspec
ts of biological evolution that remain fascinatingly mysterious. We still
have little understanding of its powers and limits: how complex is the fun
ctionality that biology can evolve? Recent work by Leslie Valiant and coll
eagues on viewing evolution as a form of computational learning from the e
nvironment has been able to obtain (based on a number of simplifying assum
ptions) certain bounds on the complexity of the genetic circuits that can
evolve within feasible quantities of time and resources. Here we will go o
ver the key concepts behind these results\, and discuss possible connectio
ns to more established notions in evolutionary biology: especially modular
ity and evolvability\, which are thought to be key organising principles o
f biological systems. In particular\, we speculate on the link between two
notions of ‘evolvability’: the evolvability of a system in terms of h
ow agile it is in responding to novel goals or environments\, and the evol
vability of certain kinds of gene network functionality in terms of its co
mputational complexity. Drawing on some recent work on the complexity of g
raph-theoretic problems on modular networks\, we suggest that modularity a
s an organising principle may have its raison d’etre in its ability to e
nhance evolvability\, in both its senses.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=3046
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3046
END:VEVENT
BEGIN:VEVENT
SUMMARY:Constructive and Non-constructive Aspects of the Lovasz Local Lemm
a
DTSTART;VALUE=DATE-TIME:20130701T090000Z
DTEND;VALUE=DATE-TIME:20130701T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3048@cern.ch
DESCRIPTION:In three talks\, I will describe aspects of the Local Lemma th
at have recently been uncovered by Moser & Tardos\, Pegden\, and David Har
ris and myself. As a running example\, we will consider the following type
of "graph transversal" problem introduced by Bollobas\, Erdos and Szemere
di in the 1970s: given a graph G=(V\,E) and an integer s\, for how small a
b can we guarantee that no matter how V has been partitioned into blocks\
, each of size at least b\, there is a way of choosing one vertex from eac
h block such that the chosen vertices do not induce a clique on s vertices
?\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3048
LOCATION:Colaba Campus A-269 (DAA Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3048
END:VEVENT
BEGIN:VEVENT
SUMMARY:Constructive and Non-constructive Aspects of the Lovasz Local Lemm
a
DTSTART;VALUE=DATE-TIME:20130702T050000Z
DTEND;VALUE=DATE-TIME:20130702T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3049@cern.ch
DESCRIPTION:In three talks\, I will describe aspects of the Local Lemma th
at have recently been uncovered by Moser & Tardos\, Pegden\, and David Har
ris and myself. As a running example\, we will consider the following type
of "graph transversal" problem introduced by Bollobas\, Erdos and Szemere
di in the 1970s: given a graph G=(V\,E) and an integer s\, for how small a
b can we guarantee that no matter how V has been partitioned into blocks\
, each of size at least b\, there is a way of choosing one vertex from eac
h block such that the chosen vertices do not induce a clique on s vertices
?\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3049
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3049
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Bayesian Approach To Stochastic Root Finding
DTSTART;VALUE=DATE-TIME:20130705T090000Z
DTEND;VALUE=DATE-TIME:20130705T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3063@cern.ch
DESCRIPTION:A stylized model of one-dimensional stochastic root-finding in
volves repeatedly querying an oracle as to whether the root lies to the le
ft or right of a given point x. The oracle answers this question\, but the
received answer is incorrect with probability 1−p(x). A Bayesian-style
algorithm for this problem that assumes knowledge of p(.) repeatedly updat
es a density giving\, in some sense\, one's belief about the location of t
he root. We demonstrate how the algorithm works\, and provide some results
that shed light on its performance\, both when p(.) is constant and when
p(.) varies with x.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=3063
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3063
END:VEVENT
BEGIN:VEVENT
SUMMARY:Asymmetric Communication Complexity and Its Application
DTSTART;VALUE=DATE-TIME:20130712T090000Z
DTEND;VALUE=DATE-TIME:20130712T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3095@cern.ch
DESCRIPTION:We consider two-party communication complexity\, the ``asymmet
ric case''\, when the input sizes of the two players differ significantly.
Most of previous work on communication complexity only considers the tota
l number of bits sent\, but we will see trade-offs between the number of b
its the first player sends and the number of bits the second sends. These
types of questions are closely related to the complexity of static data st
ructure problems in the cell probe model (which we will not discuss in thi
s talk). We will see a simple application of this in membership problem.\
n\nRef: \n(1) Peter Bro Miltersen\, Noam Nisan\, Shmuel Safra\, Avi Wigder
son: On Data Structures and Asymmetric Communication Complexity. J. Comput
. Syst. Sci. 57(1): 37-49 (1998)\n \n(2) Eyal Kushilevitz\, Noam Nisan: Co
mmunication complexity. Cambridge Univ press: 53- 56 (1997)\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=3095
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3095
END:VEVENT
BEGIN:VEVENT
SUMMARY:Unambiguity and Timed Languages: Automata\, Logics\, Expressivenes
s
DTSTART;VALUE=DATE-TIME:20130715T053000Z
DTEND;VALUE=DATE-TIME:20130715T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3072@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3072
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3072
END:VEVENT
BEGIN:VEVENT
SUMMARY: Split-width and the Verification of Concurrent Recursive Programs
DTSTART;VALUE=DATE-TIME:20130717T050000Z
DTEND;VALUE=DATE-TIME:20130717T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3098@cern.ch
DESCRIPTION:In this talk we describe a simple measure\, related to Tree/Cl
ique width\, on the structure of the runs of concurrent recursive programs
and illustrate its use in the verification of properties expressed in Mon
adic Second-order Logic. (Joint work with Aiswarya Cyriac and Paul Gastin)
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3098
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3098
END:VEVENT
BEGIN:VEVENT
SUMMARY:Leonid Gurvits's Proof of the Van der Waerden Conjecture
DTSTART;VALUE=DATE-TIME:20130719T090000Z
DTEND;VALUE=DATE-TIME:20130719T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3105@cern.ch
DESCRIPTION:The permanent of a doubly stochastic matrix is at least that o
f the matrix with each entry 1/n (the matrices are n x n). This theorem\,
popularly known as the 'Van der Waerden conjecture'\, remained open for ov
er fifty years\, before it was finally proved by Falikman (1979) and Egor
itsjev (1980). Relatively recently\, in 2008\, Leonid Gurvits gave an amaz
ingly short proof for it. In this talk\, we will discuss this new proof.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3105
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3105
END:VEVENT
BEGIN:VEVENT
SUMMARY:Unbalanced Random Matching Markets
DTSTART;VALUE=DATE-TIME:20130724T090000Z
DTEND;VALUE=DATE-TIME:20130724T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3109@cern.ch
DESCRIPTION:We analyze large random matching markets with unequal numbers
of men and women. We find that being on the short side of the market confe
rs a large advantage. For each agent\, assign a rank of 1 to the agent's m
ost preferred partner\, a rank of 2 to the next most preferred partner and
so forth. If there are n men and n+1 women then\, we show that with high
probability\, in any stable matching\, the men's average rank of their wiv
es is no more than 3logn\, whereas the women's average rank of their husba
nds is at least n/(3logn). Furthermore\, with high probability\, the fract
ion of agents with multiple stable partners is vanishing as the market gro
ws large\, i.e.\, such unbalanced random matching markets have a 'small co
re'.\n\nOur results suggest that a 'small core' may be generic in matching
markets\, contrary to prior beliefs (based on joint work with Itai Ashlag
i and Jacob Leshno).\n\nBio: Yashodhan Kanoria is Assistant Professor in t
he Decision\, Risk and Operations Division at Columbia Business School. He
obtained a B. Tech. from IIT Bombay and a PhD from Stanford\, both in Ele
ctrical Engineering. His current research interests include matching marke
ts\, graphical models and probability.\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=3109
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3109
END:VEVENT
BEGIN:VEVENT
SUMMARY:Non-asymptotic Bounds for Various Information Theoretic Problems U
sing Smooth Renyi Quantities
DTSTART;VALUE=DATE-TIME:20130726T090000Z
DTEND;VALUE=DATE-TIME:20130726T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3111@cern.ch
DESCRIPTION:In this talk we will discuss about the role played by smooth R
enyi quantities in non-asymptotic information theory. In particular\, we w
ill discuss about various source coding and channel coding problems in the
non-asymptotic regime.\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=3111
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3111
END:VEVENT
BEGIN:VEVENT
SUMMARY:Deterministic Communication Protocol for Functions with Bounded Ra
nk
DTSTART;VALUE=DATE-TIME:20130802T090000Z
DTEND;VALUE=DATE-TIME:20130802T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3119@cern.ch
DESCRIPTION:In this talk we will go through the results in the paper title
d "Communication is bounded by root of rank" by Shachar Lovett.\nThe main
result is to give a deterministic communication protocol which transfers a
t most $O(\\sqrt{r}\\log r)$ bits for computing any function $f:X\\times
Y \\rightarrow \\{0\,1\\}$ whose matrix has rank $r$. Equivalently this ca
n be stated as saying that any graph whose adjacency matrix has rank $r$ h
as chromatic number at most $2^{O(\\sqrt{r}\\log r}$. This is a major impr
ovement in the log-rank conjecture proposed by Lovasz and Saks. The proof
is based on analyzing the discrepancy of Boolean functions.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=3119
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3119
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Testing Bent Functions
DTSTART;VALUE=DATE-TIME:20130805T103000Z
DTEND;VALUE=DATE-TIME:20130805T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3124@cern.ch
DESCRIPTION:A bent function is a Boolean function all of whose Fourier coe
fficients are equal in absolute value. These functions have been extensive
ly studied in cryptography and play an important role in cryptanalysis and
design of cryptographic systems. We study bent functions in the framework
of property testing. In particular\, we show that testing whether a given
Boolean function on $n$ variables is bent\, or $\\frac{1}{8}$-far from be
ing bent\, requires $\\Omega(n^2)$ queries. As an intermediate step in our
proof\, we show that the query complexity of testing if a given function
is a quadratic bent function\, or $\\frac{1}{4}$-far from being so\, is $\
\Theta(n^2)$. We remark that this problem is equivalent to testing affine-
isomorphism to the inner product function. Our proof exploits the recent c
onnection between property testing and parity decision trees due to Chakra
borty and Kulkarni. We believe our techniques might be useful in proving l
ower bounds for other properties of quadratic polynomials.\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=3124
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3124
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Mathematical Problems Arising in the Study of Chemical Reacti
on Networks
DTSTART;VALUE=DATE-TIME:20130806T103000Z
DTEND;VALUE=DATE-TIME:20130806T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3099@cern.ch
DESCRIPTION:Networks of chemical reactions have natural underlying combina
torial structure\, allowing them to be represented as graphs or digraphs\,
perhaps with additional vertex or edge colourings/labellings. A variety o
f recent theorems demonstrate that finite computations on these graphs can
determine the allowed dynamics of these systems to a surprising extent. F
or example\, graph computations can lead to claims about the nature\, uniq
ueness and stability of limit sets\; the monotonicity of semiflows\; the p
ossibility of bifurcations\; and so forth. Thus combinatorial approaches a
re making an increasing contribution to the study of allowed behaviours in
these dynamical systems. I'll present a brief overview of some problems i
n this area\, how these intersect with certain themes in analysis\, convex
geometry and combinatorics\, but also how they suggest new mathematical q
uestions of interest in their own right.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=3099
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3099
END:VEVENT
BEGIN:VEVENT
SUMMARY:Complex Online Decision Making
DTSTART;VALUE=DATE-TIME:20130807T053000Z
DTEND;VALUE=DATE-TIME:20130807T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3125@cern.ch
DESCRIPTION:Suppose you want to find the least congested route in an ad ho
c network. Each link's rate is unknown and stochastic\, and each time you
get to see the minimum rate (i.e.\, bottleneck) along any route you pick.
More generally\, given such partial feedback\, suppose you want to find th
e maximum multicommodity flow in a network with stochastic links. A direct
approach is to treat this as a standard multi-armed bandit with #arms = #
routes\, but which leads to a potentially huge number of actions. Moreover
\, this ignores the structure of the problem -- the actual number of indep
endent parameters (i.e.\, the unknown links) is much smaller.\n\nWe model
a very general class of such "complex bandit" online problems\, and presen
t the natural\, Bayesian-inspired Thompson sampling algorithm to solve the
m. A key contribution is to provide a general\, frequentist bound on the r
egret of this algorithm which explicitly encodes the information structure
of the complex bandit. This general bound is used to give non-trivial reg
ret guarantees for a class of complex\, subset selection bandit problems.\
n\nBio: Aditya Gopalan received the Ph.D. degree in electrical engineering
at The University of Texas at Austin. He received the B.Tech. and M.Tech
. degrees in electrical engineering from the Indian Institute of Technolog
y Madras in 2006. He was a summer intern at the Corporate Research and Dev
elopment Center\, Qualcomm Inc.\, in 2009. His current research interests
include machine learning\, network algorithms and stochastic control. He i
s currently a postdoctoral researcher research fellow in the Faculty of El
ectrical Engineering at the Technion- Israel Institute of Technology\, Hai
fa\, Israel.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=3125
LOCATION:Colaba Campus A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3125
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fundamental Limits on Communication Energy Via "Information-fricti
on"
DTSTART;VALUE=DATE-TIME:20130816T083000Z
DTEND;VALUE=DATE-TIME:20130816T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3136@cern.ch
DESCRIPTION:Just as there are frictional losses in moving a weight on a su
rface\, there are also frictional losses in moving information on a substr
ate. This "information-friction" has received little attention within both
theoretical computer science and information theory. But it is important
in both! I will discuss our work at the intersection of both fields: infor
mation-frictional losses in the circuitry at the transmitter and the recei
ver of a communication system.\n\nFrom a theoretical perspective\, I will
show how accounting for these losses leads to a novel understanding of *to
tal* energy consumed in a communication system that goes beyond the transm
it-power-centric Shannon theory. For instance\, we show that approaching t
he Shannon-limit on transmit power is fundamentally accompanied with incre
asing amounts of power in the encoding/decoding circuitry (in both circuit
gates and circuit wires). Thus the total-power-minimizing strategies must
operate far from the Shannon limit.\n\nFrom a practical perspective\, our
early results -- that rely on circuit simulations for power consumption e
stimation -- show that novel total-power-optimizing strategies can lead to
substantial power reductions in short-distance communication systems. I w
ill discuss our work on two applications (i) millimeter wave wireless\; an
d (ii) data-center Ethernet.\n\nBio: Pulkit Grover (Ph.D. UC Berkeley'10\,
B.Tech.'03\, M.Tech.'05 IIT Kanpur) is an assistant professor at CMU. He
is interested in interdisciplinary research directed towards developing a
science of information for making decentralized systems (from low-power co
mmunication systems to large control systems) energy-efficient and stable.
He is the recipient of the best student paper award at the IEEE Conferenc
e in Decision and Control (CDC) 2010\; and the 2012 Leonard G. Abraham bes
t paper award from the IEEE Communications Society for his work on energy-
efficient communication. For his dissertation research\, he received the 2
011 Eli Jury Award from the Department of EECS at UC Berkeley.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=3136
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3136
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Statistical View to Boosting
DTSTART;VALUE=DATE-TIME:20130816T103000Z
DTEND;VALUE=DATE-TIME:20130816T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3140@cern.ch
DESCRIPTION:In machine learning\, AdaBoost has been an extremely popular b
oosting algorithm to improve the performance of ``weak learners". AdaBoost
was initially proposed by Schapire and Freund from an algorithmic perspec
tive. The statistical machine learners (who maintain that all machine lear
ning algorithms are derived from a statistical framework) remained skepti
cal of AdaBoost until Freidman et.al gave a statistical view of boosting a
nd proved that boosting is equivalent to fitting additive models.\n\nWe wi
ll study this generalized boosting models and obtain AdaBoost as a special
ized version.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=3140
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3140
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Markov Chain Approximation for Choice Modeling
DTSTART;VALUE=DATE-TIME:20130820T090000Z
DTEND;VALUE=DATE-TIME:20130820T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3127@cern.ch
DESCRIPTION:Assortment planning is an important problem that arises in man
y industries such as retailing and airlines. One of the key challenges in
an assortment planning problem is to identify the ``right model'' for the
substitution behavior of customers from the data. Error in model selection
can lead to highly sub-optimal decisions. In this paper\, we present a ne
w choice model that is a simultaneous approximation for all random utility
based discrete choice models including the multinomial logit\, the nested
logit and mixtures of multinomial logit models. Our model is based on a n
ew primitive for substitution behavior where substitution from one product
to another is modeled as a state transition of a Markov chain.\n \nWe sho
w that the choice probabilities computed by our model are a good approxima
tion to the true choice probabilities of any random utility discrete based
choice model under mild conditions. Moreover\, they are exact if the unde
rlying model is a Multinomial logit model. We also give a procedure to est
imate the parameters of the Markov chain model that does not require any k
nowledge of the latent choice model. Furthermore\, we show that the assort
ment optimization problem under our choice model can be solved efficiently
in polynomial time. This is quite surprising as we can not even express t
he choice probabilities using a functional form. Our numerical experiments
show that the average maximum relative error between the estimates of the
Markov chain choice probability and the true choice probability is less t
han 3% (the average being taken over different offer sets). Therefore\, ou
r model provides a tractable data-driven approach to choice modeling and a
ssortment optimization that is robust to model selection errors.\n\nhttps:
//indico.tifr.res.in/indico/conferenceDisplay.py?confId=3127
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3127
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimal Change of Measure for Model Selection and Efficient Simula
tion of Large Deviation Probability with Financial Applications
DTSTART;VALUE=DATE-TIME:20130823T090000Z
DTEND;VALUE=DATE-TIME:20130823T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3149@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3149
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3149
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Johnson-Lindenstrauss Lemma With Independent Subgaussian Project
ion Coefficients
DTSTART;VALUE=DATE-TIME:20130823T103000Z
DTEND;VALUE=DATE-TIME:20130823T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3158@cern.ch
DESCRIPTION:The Johnson-Lindenstrauss lemma asserts that any n-point set i
n any Euclidean space can be mapped to a Euclidean space of dimension $k=
O(\\epsilon^{-2} \\log n)$ so that all distances are preserved upto a mult
iplicative factor between $(1 - \\epsilon)$ and $(1+\\epsilon)$. There are
several proofs of $JL Lemma$\, the notable ones being by Indyk and Motwan
i\, Dasgupta and Gupta\, Achlioptas. All these proofs obtain such a mappin
g as a linear map $R^n -> R^k$ with a suitable random matrix. In this tal
k I intend to present a portion of a 2008 paper by Matoušek\, where the a
uthor gives a simple and self-contained proof of the $JL-lemma$ which subs
umes several of the earlier proofs. The paper uses a distribution on $n$ b
y $k$ matrices where the entries are independent random variables with mea
n $0$ and bounded variance\, and with a uniform subgaussian tail. The dist
ributions used by Indyk-Motwani and Achlioptas turn out to both fall in th
is category.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=3158
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3158
END:VEVENT
BEGIN:VEVENT
SUMMARY:Static vs Adjustable Solutions in Dynamic Optimization
DTSTART;VALUE=DATE-TIME:20130828T113000Z
DTEND;VALUE=DATE-TIME:20130828T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3128@cern.ch
DESCRIPTION:We study the performance of static solutions for two-stage adj
ustable robust linear optimization problems with uncertain constraint and
objective coefficients and give a tight characterization of the adaptivity
gap. Computing an optimal adjustable robust optimization problem is often
intractable\, but a static solution can be computed efficiently in most c
ases. We show that for a fairly general class of uncertainty sets\, a stat
ic solution is optimal for the two-stage adjustable robust linear optimiza
tion problem. Furthermore\, when a static solution is not optimal\, we giv
e a tight approximation bound on the performance of the static solution th
at is related to a measure of non-convexity of a transformation of the unc
ertainty set. We also show that our bound is at least as good (and in many
case significantly better) as the bound given by the symmetry of the unce
rtainty set.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=3128
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3128
END:VEVENT
BEGIN:VEVENT
SUMMARY:Randomized Communication Complexity of Set Disjointness
DTSTART;VALUE=DATE-TIME:20130830T103000Z
DTEND;VALUE=DATE-TIME:20130830T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3171@cern.ch
DESCRIPTION:In this talk we will study the communication complexity of the
disjointness function\, in which each of two players holds a $k$-subset o
f a universe of size $n$ and the goal is to determine whether the sets are
disjoint. In the model of a common random string we prove that $O(k)$ com
munication bits are sufïcient\, regardless of $n$. In the model of priv
ate random coins $O(k +\\ log \\log n)$ bits sufïce. Both results are a
symptotically tight.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=3171
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3171
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum Information Processing Using Spin Coherent States
DTSTART;VALUE=DATE-TIME:20130905T090000Z
DTEND;VALUE=DATE-TIME:20130905T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3164@cern.ch
DESCRIPTION:Spin coherent states are the matter equivalent of optical cohe
rent states\, where a large number of spins occupy the same quantum state.
Experiments have now advanced to the point that such states\, as realize
d in atomic Bose-Einstein condensates (BECs)\, can be coherently controlle
d with long decoherence times. Here we discuss a method of using spin coh
erent states\, particularly of the BEC form\, directly for quantum informa
tion processing. We construct a general framework for quantum algorithms
to be executed using spin coherent states. We illustrate the scheme by an
application to quantum algorithms and discuss the effects of decoherence
induced by the large number of particles in the BEC. We also discuss a ne
w quantum teleportation protocol that allows for the transfer of a spin co
herent states between two parties\, assisted by entanglement. In the prot
ocol\, two macroscopic spin state are entangled\, and an initially unknown
spin coherent state is transferred to a distant location. Unlike standar
d quantum teleportation where a single qubit is transferred\, a macroscopi
c ensemble of spins is teleported in our scheme. By the use of a special
class of entangled states\, it is possible to avoid the detrimental effect
s of decoherence on such macroscopic state on the protocol.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=3164
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3164
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mean Field Games with Substitutable Goods
DTSTART;VALUE=DATE-TIME:20130905T103000Z
DTEND;VALUE=DATE-TIME:20130905T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3173@cern.ch
DESCRIPTION:We study continuous time Bertrand competition in which a large
number of firms producing similar goods compete with one another by setti
ng prices. Interactions are of mean field type in the sense that the deman
d faced by a producer is affected by the others through their average pric
e. We consider the setting of dynamic game with uncertain market demand\,
where firms of different sizes compete on prices and produce to clear the
market demand until they have depleted their lifetime capacity. We set up
the nonzero-sum stochastic differential game of mean field type and its as
sociated forward/backward system of partial differential equations in the
case of linear demand functions. Asymptotic approximation enables us to de
duce certain qualitative features of the game in the limit of small compet
ition. The equilibrium of the game is further studied using numerical solu
tions. We find that\, in accordance with the 2-player game\, firms slow do
wn production with a large degree of substitutability.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=3173
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3173
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Art Gallery Approach To Ensure That Landmarks Are Distinguishab
le
DTSTART;VALUE=DATE-TIME:20130906T103000Z
DTEND;VALUE=DATE-TIME:20130906T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3178@cern.ch
DESCRIPTION:In a 2-dimensional setting\, how many different classes of par
tially distinguishable landmarks are needed to ensure that a robot can alw
ays see a landmark without simultaneously seeing two of the same class? In
other words\, what is the minimum number of colors required to color any
guard set (not necessarily a minimal guard set) of a polygon such that no
two guards of the same colors see a common point (conflict-free coloring)?
In this talk we will see lower bounds of colors for conflict free guardin
g of general\, monotone and orthogonal polygons. This is recent work by L
awrence Erickson and Steven LaValle.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=3178
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3178
END:VEVENT
BEGIN:VEVENT
SUMMARY:Public vs Private-Coin Information Complexity
DTSTART;VALUE=DATE-TIME:20130910T090000Z
DTEND;VALUE=DATE-TIME:20130910T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3183@cern.ch
DESCRIPTION:We study the relation between public and private-coin informat
ion complexity. Improving a recent result by Brody et al.\, we prove that
a one-round private-coin protocol with information cost I can be simulated
by a one-round public-coin protocol with information cost \\le I + \\log(
I) + O(1). This question is connected to the question of compression of in
teractive protocols and direct sum for communication complexity.\n\nWe als
o give a lower bound. We exhibit a one-round private-coin protocol with in
formation cost ~ n/2 - \\log(n) which cannot be simulated by any public-co
in protocol with information cost n/2 - O(1). This example also explains a
n additive \\log factor incurred in the study of communication complexity
of correlations by Harsha et al.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=3183
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3183
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Growing Trees
DTSTART;VALUE=DATE-TIME:20130913T103000Z
DTEND;VALUE=DATE-TIME:20130913T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3192@cern.ch
DESCRIPTION:As kids\, we all have learnt that by planting a seed in the gr
ound and watering it regularly we can grow a tree. Of course there are oth
er ways to grow trees. Here we won't talk about trees that give us oxygen.
Instead\, we will present special kinds of trees known as "Decision Trees
" that are generated from data and can be used as predictive models for cl
assification and regression problems. Firstly\, we will define decision tr
ee learning. Then we will see the Classification And Regression Tree (CART
) algorithm for growing (splitting strategies) decision trees due to Briem
an et. al.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=3192
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3192
END:VEVENT
BEGIN:VEVENT
SUMMARY:Combinatorial Nullstellensatz and Some Applications
DTSTART;VALUE=DATE-TIME:20130920T103000Z
DTEND;VALUE=DATE-TIME:20130920T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3211@cern.ch
DESCRIPTION:In this talk\, we will prove two basic theorems known as the c
ombinatorial nullstellensatz. These theorems\, a substantial part of the p
olynomial method\, have found many applications in combinatorial number th
eory\, in graph theory and in combinatorics. In this talk\, we will see so
me of them.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=3211
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3211
END:VEVENT
BEGIN:VEVENT
SUMMARY:Isotonic Regression
DTSTART;VALUE=DATE-TIME:20130927T103000Z
DTEND;VALUE=DATE-TIME:20130927T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3237@cern.ch
DESCRIPTION:In statistical analysis\, regression is a widely used tool to
get a functional relationship between observations of a response variable
Y and covariate X. The goal is to estimate the regression function $$ m(X
)=E[Y|X] $$. The estimation of m(·) is done by least-squares\, where m(·
) is an arbitrary function. In isotonic regression we restrict the functio
n m(·) to be monotone. In this talk\, we will look at a geometrical meth
od to find m(·) and prove its optimality. This method leads to fast algo
rithm called Pooled Adjacent Value Algorithm (PAVA) which is widely used i
n statistical packages.\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=3237
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3237
END:VEVENT
BEGIN:VEVENT
SUMMARY:Algebraic and Logical Approaches to Circuit Complexity
DTSTART;VALUE=DATE-TIME:20131001T103000Z
DTEND;VALUE=DATE-TIME:20131001T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3225@cern.ch
DESCRIPTION:One of the outstanding open questions in computational complex
ity concerns the power of circuits that include modular counting gates alo
ng with the usual AND and OR gates. More precisely\, the class ACC consist
s of languages recognized by constant-‐depth polynomial-‐size fami
lies of circuits with unbounded fan-‐in AND\, OR and MOD-q gates for a
fixed modulus q\; the question is whether ACC is strictly contained in NC
1. At present this is known only for prime power moduli\, a result proved
in the mid 1980’s.\n\nThere are some very striking connections between t
his question and problems and techniques originating in both finite model
theory and the algebraic theory of finite automata. This talk is an overvi
ew of these connections along with the efforts\, some by now quite old\, s
ome much more recent\, to exploit them in addressing the status ofACC.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3225
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3225
END:VEVENT
BEGIN:VEVENT
SUMMARY:Surreal Numbers
DTSTART;VALUE=DATE-TIME:20131004T103000Z
DTEND;VALUE=DATE-TIME:20131004T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3246@cern.ch
DESCRIPTION:In past there were many attempts to axiomatically define Numbe
r System. Piano’s\, Dedekind’s\, Cantor’s and Von Newman’s System
are few examples among popular approaches. The definition and constructio
n of Surreal Numbers is another such attempt by Prof. J H Conway. However
the name is introduced by D. E. Knuth due to his book titled “Surreal N
umbers”. In this talk we will start by defining Numbers and “\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3246
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3246
END:VEVENT
BEGIN:VEVENT
SUMMARY:Schmidt Games
DTSTART;VALUE=DATE-TIME:20131011T103000Z
DTEND;VALUE=DATE-TIME:20131011T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3252@cern.ch
DESCRIPTION:This talk aims to discuss a class of infinite games discovered
by Wolfgang M. Schmidt (1966) and the way it has been used to establish t
he largeness of certain subsets of Rn. We only assume the definition of cl
osed balls in Euclidean spaces.\n\nReference: Wolfgang M. Schmidt\, On Bad
ly Approximable numbers and certain games\, Trans. Amer. Math. Soc. 123 (1
966)\, 178-199\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=3252
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3252
END:VEVENT
BEGIN:VEVENT
SUMMARY:Privacy-preserving Algorithms for Signal Processing and Machine Le
arning
DTSTART;VALUE=DATE-TIME:20131017T103000Z
DTEND;VALUE=DATE-TIME:20131017T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3196@cern.ch
DESCRIPTION:Large-scale data mining of private information (such as medica
l or financial records) has the potential to revolutionize the way we live
. On the positive side\, it can enable large-scale studies on the comparat
ive effectiveness of treatments\, novel risk factors for diseases\, and pa
tient-centered personalized medicine. There are many challenges to overco
me in order to realize these benefits\, ranging from the way we represent
data to legal and policy arrangements between medical institutions. Uniti
ng these is a concern about the privacy -- we want to guarantee low privac
y risk and high precision\, or utility. The fundamental question to answe
r is this : how much data do we need to guarantee acceptable levels privac
y and utility?\n\nIn this talk I describe practical approaches for managin
g this tradeoff. These methods guarantee differential privacy\, a cryptog
raphically-motivated definition of privacy which has been widely adopted i
n the computer science community. I will discuss basic ideas from differe
ntial privacy and how to use them to build algorithms for classification a
nd dimensionality reduction\, which are two of the most common tasks in ma
chine learning. I will also describe some exciting future prospects for p
rivacy-preserving algorithms in signal processing\, optimization\, and lea
rning (joint work with Kamalika Chaudhuri (UCSD)\, Claire Monteleoni (GWU)
\, Kaushik Sinha (Wichita State U)\, and Shuang Song (UCSD)).\n\nBio: Anan
d Sarwate is currently a Research Assistant Professor at the Toyota Techno
logical Institute at Chicago\, a philanthropically endowed academic instit
ute located on the University of Chicago campus. Prior to that he was a po
stdoc in the Information Theory and Applications Center (ITA) at UC San Di
ego. He received his PhD from UC Berkeley in 2008\, and undergraduate deg
rees in Mathematics and Electrical Engineering from MIT in 2002. He recei
ved the Demetri Angelakos and Samuel Silver awards from the EECS departmen
t at UC Berkeley. He is broadly interested in algorithms applied to probl
ems in distributed systems\, signal processing\, machine learning\, statis
tics\, and privacy and security. He will be joining the Department of Ele
ctrical and Computer Engineering as an Assistant Professor in January 2014
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3196
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3196
END:VEVENT
BEGIN:VEVENT
SUMMARY:Streaming Algorithm
DTSTART;VALUE=DATE-TIME:20131018T103000Z
DTEND;VALUE=DATE-TIME:20131018T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3263@cern.ch
DESCRIPTION:In the streaming model\, we look at a sequence of elements (a_
1\,a_2\,...\,a_m) where each element comes from the set {1\,2\,...\,n}. Ou
r aim is to compute some function on these numbers\, but using very "littl
e" space. The best algorithm would be one which uses O(log m + log n) spac
e.In the course of the talk we will show that finding the k^{th} highest e
lement in a stream can be done better than O(k log n) bits. We will also l
ook at the space complexity of frequency moments (this work is by Alon\,Ma
tias\, Szegedy for which they won the Godel prize).\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=3263
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3263
END:VEVENT
BEGIN:VEVENT
SUMMARY:Signal Processing for Systems With Low Precision Quantization
DTSTART;VALUE=DATE-TIME:20131023T040000Z
DTEND;VALUE=DATE-TIME:20131023T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3255@cern.ch
DESCRIPTION:In system design\, digital signal processing involves operatio
n upon and analysis of digital signals. The real physical signals are temp
orally or spatially varying physical quantities (for example\, sound\, ele
ctromagnetic radiation\, images etc.) which can take values over an infini
te\, uncountable set. Digitizing physical signals is achieved in two steps
sampling in the temporal or spatial domain followed by quantizing the sam
ples to a finite set of values. For a bandlimited physical signal taking v
alues over an infinite set\, sampling\, if done over a sufficiently fine g
rid\, can be achieved with no loss of information. However\, quantization
is inherently a non-linear and lossy process. If the quantization precisio
n is sufficiently large\, the non-linearities can be ignored and signal pr
ocessing algorithms for the system can be designed assuming no quantizatio
n i.e. samples are represented by their actual values. However\, if the pr
ecision is low\, the non-linearities become significant and it becomes nec
essary to take into account these effects while designing signal processin
g algorithms for such systems. In this thesis\, we consider two such appli
cations which require designing signal processing algorithms for low preci
sion systems\, viz. transceiver design for multiGigabit communication syst
ems and mask design in optical lithography.\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=3255
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3255
END:VEVENT
BEGIN:VEVENT
SUMMARY:PAC Subset Selection in Stochastic Multi-armed Bandits
DTSTART;VALUE=DATE-TIME:20131101T053000Z
DTEND;VALUE=DATE-TIME:20131101T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3208@cern.ch
DESCRIPTION:We consider the problem of selecting\, from among n real-value
d random variables\, a subset of size m of those with the highest means\,
based on efficiently sampling the random variables. This problem\, which w
e denote Explore-m\, finds application in a variety of areas\, such as sto
chastic optimization\, simulation and industrial engineering\, and on-line
advertising. The theoretical basis of our work is an extension of a previ
ous formulation using multi-armed bandits that is devoted to identifying j
ust the single best of n random variables (Explore-1). Under a PAC setting
\, we provide algorithms for Explore-m and bound their sample complexity.\
n\nOur main contribution is the $LUCB$ algorithm\, which\, interestingly\,
bears a close resemblance to the well-known $UCB$ algorithm for regret mi
nimization. We derive an expected sample complexity bound for $LUCB$ that
is novel even for single-arm selection. We then improve the problem-depen
dent constant in this bound through a novel algorithmic variant called $KL
-LUCB$. Experiments affirm the relative efficiency of $KL-LUCB$ over othe
r algorithms for Explore-m. Our contributions also include a lower bound o
n the worst case sample complexity of such algorithms.\n\nBio: Shivaram Ka
lyanakrishnan is a scientist at Yahoo Labs Bangalore. His primary research
interests lie in the fields of artificial intelligence and machine learni
ng\, spanning areas such as reinforcement learning\, agents and multiagent
systems\, humanoid robotics\, multi-armed bandits\, and on-line advertisi
ng. He obtained his Ph.D. in Computer Science from the University of Texas
at Austin (2011)\, and his B.Tech. in Computer Science and Engineering fr
om the Indian Institute of Technology Madras (2004). He has extensively us
ed robot soccer as a test domain for his research\, and has actively contr
ibuted to initiatives such as RoboCup and the Reinforcement Learning compe
titions.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3208
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3208
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimization in Very High Dimensions
DTSTART;VALUE=DATE-TIME:20131101T103000Z
DTEND;VALUE=DATE-TIME:20131101T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3276@cern.ch
DESCRIPTION:Optimization problems where the number of variables $n$ is $10
^4$ or more are now ubiquitous in all big data applications such as machin
e learning\, social network science\, signal processing\, etc. For such pr
oblems\, even basic methods such as the steepest descent and conjugate gra
dient are not favored because of their $O(n^2)$ or more cost per iteration
. Inspired by Kaczmarz\, this article proposes a simple framework to devel
op randomized descent methods for optimization in very high dimensions. In
each iteration\, the new estimate is the optimum of the objective functio
n restricted to a random $k-$dimensional affine space passing through the
old estimate. If these affine spaces are chosen cleverly enough\, then one
can simultaneously have a) cheap iterations\, b) fast convergence and c)
the ability to tradeoff between the two by varying $k.$ To demonstrate the
efficacy of this framework\, we present here a novel algorithm to solve q
uadratic programs. The per iteration cost of this algorithm is $O(nk) + O(
k^3)$ and its expected convergence rate is $(1 - \\Omega(\\frac{k}{n})).$\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3276
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3276
END:VEVENT
BEGIN:VEVENT
SUMMARY:Model Reduction in Systems Biology
DTSTART;VALUE=DATE-TIME:20131106T103000Z
DTEND;VALUE=DATE-TIME:20131106T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3262@cern.ch
DESCRIPTION:Models based on Ordinary Differential Equations (ODEs) or thei
r stochastic counterparts are routinely applied in systems biology to unde
rstand\, analyze and predict properties and behaviour of biomolecular dyna
mical systems\, such as enzymatic networks. However\, the system's propert
ies and behaviour depend strongly on the modeling equations and the assump
tions we make about the system. How robust are predictions with respect to
the assumptions? What can we say about the mathematical relationship betw
een different models of the same biological system?\n\nIn recent years\, m
odel reduction -- how do we simplify a complex model without loosing essen
tial properties of the model -- has been a topic in stochastic and determi
nistic reaction network theory. In this talk\, I will discuss model reduct
ion and some recent results on model reduction\, based on deterministic as
well as stochastic systems\, at steady state\, as well as out of steady s
tate.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=326
2
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3262
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Power of Conditional Samples in Distribution Testing
DTSTART;VALUE=DATE-TIME:20131107T103000Z
DTEND;VALUE=DATE-TIME:20131107T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3280@cern.ch
DESCRIPTION:We define and examine the power of the conditional-sampling or
acle in the context of distribution-property testing. The conditional-samp
ling oracle for a discrete distribution $\\mu$ takes as input a subset $S$
of the domain\, and outputs a random sample $i\\in S$ drawn according to
$\\mu$\, conditioned on $S$ (and independently of all prior samples). The
conditional-sampling oracle is a natural generalization of the ordinary sa
mpling oracle in which $S$ always equals the whole domain.\n\nWe show that
with the conditional-sampling oracle\, testing uniformity\, testing ident
ity to a known distribution\, and testing any label-invariant property of
distributions is easier than with the ordinary sampling oracle. On the oth
er hand\, we also show that for some distribution properties the sample-co
mplexity remains near-maximal even with conditional sampling.\n\nOne can u
se conditional sampling for various real life problems also (this is a joi
nt work with Eldar Fischer\, Yonatan Goldhirsh and Arie Matsliah).\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3280
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3280
END:VEVENT
BEGIN:VEVENT
SUMMARY:Independent Sets in $K_3^n$
DTSTART;VALUE=DATE-TIME:20131108T103000Z
DTEND;VALUE=DATE-TIME:20131108T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3288@cern.ch
DESCRIPTION:We will prove the following fact: Assume that there are n thre
e position switches that control whether the color of a single bulb is re
d\, green or yellow. You are told that whenever you change the position of
all the switches then the color of the light changes. Prove that in fact
the bulb is controlled by only one of the switches.\n\nRef: Graph Products
\, Fourier Analysis and Spectral Techniques by Alon\, Dinur\, Friedgut\, S
udakov (http://link.springer.com/article/10.1007%2Fs00039-004-0478-3)\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3288
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3288
END:VEVENT
BEGIN:VEVENT
SUMMARY:Impact of Physical Layer Re-transmissions on User Experience
DTSTART;VALUE=DATE-TIME:20131113T083000Z
DTEND;VALUE=DATE-TIME:20131113T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3279@cern.ch
DESCRIPTION:There is an ever increasing demand for instant and reliable in
formation on modern communication networks. Physical layer re-transmission
s can be thought of as opportunistic coding schemes. However\, this result
s in non-deterministic delay added to packet transmissions. In this talk\,
I first present the impact of physical layer decisions on transmit buffer
occupancy. Next\, I present the impact of physical layer re-transmissions
on end-user experience.\n\nBio: Parimal Parag received B. Tech. and M. Te
ch. degrees from IIT Madras in 2004\; and the PhD degree from Texas A&M Un
iversity in 2011. He was at Stanford University and Los Alamos National La
boratory\, in autumn of 2010 and summer of 2007\, respectively. He is curr
ently a senior systems engineer at ASSIA Inc.\, working on performance eva
luation\, monitoring\, and control of large broadband communication networ
ks. His research interests include application of stochastic and optimizat
ion methods to communication networks. His other research interests lie in
the areas of statistical signal processing\, queueing theory\, informatio
n theory\, combinatorics\, and probability theory.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=3279
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3279
END:VEVENT
BEGIN:VEVENT
SUMMARY:Cheating Husbands - A Study of the Interplay Between Knowledge\, A
ction\, and Communication
DTSTART;VALUE=DATE-TIME:20131115T103000Z
DTEND;VALUE=DATE-TIME:20131115T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3298@cern.ch
DESCRIPTION:The relationship between knowledge and action is a fundamental
one: a processor in a computer network (or a robot or a person\, for that
matter) should base its actions on the knowledge (or information) it has.
One of the main uses of communication is passing around information that
may eventually be required by the receiver in order to decide upon subsequ
ent actions. Understanding the relationship between knowledge\, action\, a
nd communication is fundamental to the design of computer network protocol
s\, intelligent robots\, etc. By looking at a number of variants of the ch
eating husbands puzzle\, we illustrate the subtle relationship between kno
wledge\, communication\, and action in a distributed environment.\n\nRefe
rence::"Cheating husbands and other stories: A case study of knowledge\, a
ction\, and communication" - Yoram Moses\, Danny Dolev\, Joseph Y. Halper
n [Distributed Computing (1986)\, Volume 1\, Issue 3\, pp 167-176 ]\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3298
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3298
END:VEVENT
BEGIN:VEVENT
SUMMARY:Price of Anarchy\, Auctions\, and Approximations
DTSTART;VALUE=DATE-TIME:20131119T103000Z
DTEND;VALUE=DATE-TIME:20131119T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3281@cern.ch
DESCRIPTION:In many applications driven by the internet\, computation has
to be performed in the presence of strategic interactions between multiple
autonomous entities. This raises the question: What happens when the inpu
t to an algorithm are controlled by rational agents\, and each of them\, i
n turn\, is affected by the algorithm's output? A systematic study of this
question led to the growth of ``algorithmic game theory''. In the talk I
will focus on two important topics within this field -- the study of auct
ions\, and price of anarchy.\n\nMotivated by the electronic markets such a
s eBay and Google adwords\, I will first visit the topic of Auction Theory
from an algorithmic perspective. In this setting\, an auctioneer wants t
o sell some items to a group of bidders. The bidders' valuations for the i
tems are private knowledge\, but they are drawn from publicly known prior
distributions. An auction scheme is called ``truthful'' if it gives an inc
entive to every bidder to report her true value. The goal is to find a tru
thful auction that maximizes the seller's revenue.\n\nNaturally\, to execu
te the optimal auction we need to know the prior distributions. An intrigu
ing question is to design a truthful auction that does not require the kno
wledge of these priors\, but still manages to get a constant fraction of t
he optimal revenue. Such auctions are called ``prior-free''. I will conclu
de the first part of the talk by presenting my work on prior-free auctions
with asymmetric bidders.\n\nNext\, I will consider the game-theoretic var
iant of a classical scheduling-problem. The setting captures a distributed
environment with heterogeneous machines (or data centers) and jobs (or cl
ients) that are situated across different geographical locations. The obj
ective is to minimize the weighted sum of the completion times of the jobs
. Each job\, however\, is a selfish agent and selects a machine that mini
mizes its own completion time. This defines a game between the jobs. A N
ash equilibrium of this game is an outcome where no job wants to switch to
another machine. The ``price of anarchy'' is the maximum possible ratio b
etween the objective at a Nash equilibrium and the objective at the global
ly-optimal outcome. Intuitively\, it measures the degradation in overall s
ystem-performance due to the presence of selfish jobs.\n\nI will present a
generic characterization of the scheduling policies which give a small Pr
ice of Anarchy. On the technical side\, I will show how to derive this res
ult using a linear program for the underlying optimization problem and dua
l-fitting.\n\nI will conclude the talk by presenting some interesting open
directions for future research.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=3281
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3281
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Finding a Set of Healthy Individuals From a Large Population
DTSTART;VALUE=DATE-TIME:20131119T053000Z
DTEND;VALUE=DATE-TIME:20131119T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3297@cern.ch
DESCRIPTION:In this talk\, we consider the problem of finding a given numb
er of "healthy" items from a large population containing a small number of
"defective" items using nonadaptive group testing. Such problems arise\,
for example\, in the spectrum hole search problem for the cognitive radio
(CR) networks. It is well-known that the primary user occupancy (active se
t) is sparse in the frequency domain over a wideband of interest. To setup
a CR network\, the secondary users need to find an appropriately wide uno
ccupied (inactive) frequency band. Thus\, the main interest here is the id
entification of only a sub-band out of the total available unoccupied band
\, i.e.\, it is an inactive subset recovery problem. Another example is a
product manufacturing plant\, where a small shipment of non-defective (ina
ctive) items has to be delivered at high priority. Once again\, the intere
st here is in the identification of a subset of the non-defective items us
ing as few tests as possible.\n\nFor this problem\, we present mutual info
rmation based upper and lower bounds on the number of nonadaptive grouptes
ts required to identify a given number of healthy items. We show that an i
mpressive reduction in the number of testsis achievable compared to the ap
proach of first identifying all the defective items and then picking the r
equired number of healthy items from the complement set. Further\, in the
nonadaptive group testing setup\, we obtain upper and lower bounds on the
number of tests under both dilution and additive noise models\, and show t
hat the bounds are order-wise tight. Our results are derived using a gener
al sparse signal model\, by virtue of which\, they are also applicable to
other important sparse signal based applications such as compressive sensi
ng (this is joint work with Abhay Sharma).\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=3297
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3297
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stochastic Blockmodels of Mixed Membership for Relational Data
DTSTART;VALUE=DATE-TIME:20131122T103000Z
DTEND;VALUE=DATE-TIME:20131122T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3308@cern.ch
DESCRIPTION:In this talk I will introduce a class of latent variable model
s for relational data. Consider the relational data of pairwise measuremen
ts\, such as presence or absence of links between a pair of objects. These
type of data arises in various environments\, for example\, Web connects
pages by links\, collection of author-recipient Emails\, scientific litera
ture connects papers by citations\, and Social Networks. These models pro
vide exploratory tools for scientific analysis in applications where the o
bservations can be represented as a collection of unipartite graphs. We wi
ll also look at a general variational inference algorithm for approximatin
g the intractable posterior distribution.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=3308
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3308
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Complete Axiom System for the Algebra of Regular Sets
DTSTART;VALUE=DATE-TIME:20131129T103000Z
DTEND;VALUE=DATE-TIME:20131129T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3323@cern.ch
DESCRIPTION:A formal axiom system for the algebraic manipulation of regula
r expressions will be discussed. This system is consistent and complete\;
equations derivable within this system are precisely the ones that are tru
e. The completeness proof is based on the uniqueness of solution of certai
n equations of regular expressions.\n\nhttps://indico.tifr.res.in/indico/c
onferenceDisplay.py?confId=3323
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3323
END:VEVENT
BEGIN:VEVENT
SUMMARY:Matchings\, Random Walks\, and Sampling
DTSTART;VALUE=DATE-TIME:20131218T083000Z
DTEND;VALUE=DATE-TIME:20131218T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3335@cern.ch
DESCRIPTION:The maximum matching problem is among the most well-studied pr
oblems in combinatorial optimization with many applications. In this talk\
, we will describe some results that illustrate surprising effectiveness o
f randomization in solving exact and approximate matching problems in smal
l space or time. Specifically\, the first part of the talk will focus on t
he problem of finding a perfect matching in a regular bipartite graph. We
will show that sampling and random walks can be used to obtain surprisingl
y fast sublinear time exact algorithms for this problem. Our approach also
yields an efficient algorithm for computing the Birkhoff-von-Neumann deco
mposition of a doubly-stochastic matrix as well as a near-linear time algo
rithm for rounding a fractional flow solution. In the second part of the t
alk\, we will consider the problem of estimating the size of a maximum mat
ching using small space in the streaming model. We show that if the edges
of the graph are streamed in a random order\, then poly-logarithmic space
suffices to estimate the maximum matching size to within a poly-logarithmi
c factor. This is based on joint works with Ashish Goel\, Michael Kapralov
\, and Madhu Sudan.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=3335
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3335
END:VEVENT
BEGIN:VEVENT
SUMMARY:From Two-Way to One-Way Finite State Transducers
DTSTART;VALUE=DATE-TIME:20131220T103000Z
DTEND;VALUE=DATE-TIME:20131220T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3354@cern.ch
DESCRIPTION:Any two-way finite state automaton is equivalent to some one-w
ay finite state automaton. This well-known result\, shown by Rabin and Sco
tt and independently by Shepherdson\, states that two-way finite state aut
omata (even non-deterministic) characterize the class of regular languages
. It is also known that this result does not extend to finite string trans
ductions: (deterministic) two-way finite state transducers strictly extend
the expressive power of (functional) one-way transducers. In particular d
eterministic two-way transducers capture exactly the class of MSO-transduc
tions of finite strings. In this talk\, we address the following definabil
ity problem: given a function defined by a two-way finite state transducer
\, is it definable by a one-way finite state transducer? By extending Rabi
n and Scott's proof to transductions\, we show that this problem is decida
ble. Our procedure builds a one-way transducer\, which is equivalent to th
e two-way transducer\, whenever one exists.\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=3354
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3354
END:VEVENT
BEGIN:VEVENT
SUMMARY:Efficiency of a Two-stage Market for Capacity Allocation
DTSTART;VALUE=DATE-TIME:20131224T060000Z
DTEND;VALUE=DATE-TIME:20131224T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3343@cern.ch
DESCRIPTION:Forward markets allow buyers and sellers of resources like net
work bandwidth or power to plan their consumption and production respectiv
ely. However\, random events (e.g. weather affecting power demand\, news e
vents compelling users to use a communication network) can happen after th
e market and before consumption that affects demand and or production of t
he resource. A spot market that happens contemporaneously with production
and consumption has the advantage that market participants act after the o
utcomes of such random events are known. A two-stage forward-spot market\,
has the potential of enjoying the benefits of both kinds of market allowi
ng participants to plan in advance while also responding to random shocks.
However\, a two-stage market might introduce the possibility for a strate
gic player to manipulate the market by creating an arbitrage between the t
wo stages\, leading to inefficiency. In this talk\, we investigate how the
efficiency of two stage markets compares with a single stage market with
price anticipating users. We show that the fundamental efficiency limit of
a two-stage market for a fixed\, divisible resource and buyers with linea
r utility functions can be no worse than 2√2−2 ≈ 82.8%. This compare
s to 75% worst-case efficiency previously known for single stage markets.\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3343
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3343
END:VEVENT
BEGIN:VEVENT
SUMMARY:DNA Sequencing and Information Theory
DTSTART;VALUE=DATE-TIME:20131227T103000Z
DTEND;VALUE=DATE-TIME:20131227T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3368@cern.ch
DESCRIPTION:DNA sequencing is an important method in Modern Biology. The p
redominant technique used is Shotgun sequencing where randomly located fra
gments (base pairs) called 'reads' are extracted from a DNA sequence. Thes
e 'reads' are later 'stitched' to reconstruct the original sequence. The m
inimum number of 'reads' required to 'stitch' the sequence reliably is an
important quantity.\n\nIn this talk\, I will show an analogy between Shotg
un sequencing and Shannon's communication model\, and discuss 'sequencing
capacity'. This is the maximum number of DNA base pairs that can be resolv
ed reliably per 'read'\, and also a fundamental limit to the performance o
f a 'stitching' algorithm. I will derive the sequencing capacity for a sim
ple model of shotgun sequencing.\n\n\nReferences :\n1. S. Motahari\, G. Br
esler\, and D. Tse\, “Information theory of DNA sequencing\,” http://a
rxiv.org/abs/1203.6233 \, 2012.\n2. S. Motahari\, G. Bresler\, and D. Tse\
, “Information Theory for DNA Sequencing: Part 1: A Basic Model\,”Proc
. IEEE International Symposium on Information Theory\, pp. 2741–2745\
, Cambridge\, MA\, July 2012\n3. J. Miller\, S. Koren\, and G. Sutton\,
“Assembly algorithms for next-generation sequencing data\,” Genomics\,
vol. 95\, pp. 315–327\, 2010.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=3368
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3368
END:VEVENT
BEGIN:VEVENT
SUMMARY:Strategizing in Communication
DTSTART;VALUE=DATE-TIME:20131230T060000Z
DTEND;VALUE=DATE-TIME:20131230T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3355@cern.ch
DESCRIPTION:We tend to think of communication in terms of the passing of i
nformation. When we say something to someone\, we change their knowledge
state as well as their beliefs. The celebrated AGM theory (Alchourron\, G
ardenfors and Makinson) as well as Dynamic Epistemic Logic follow this pat
tern.\n\nBut an alternate way to look at communication is as a game theore
tic move. We say something to someone because we want them to do somethin
g\, and the other may guess that we have strategic motives depending on ou
r own payoffs and not necessarily hers. She may then compensate for our i
nterests and take what we say with a grain of salt. We on our part may ta
ke the salt into account in what we choose to say.\n\nA conversation betwe
en a shopkeeper and a customer can take this form. Both want a sale to ta
ke place but disagree on the terms. A certain bluffing by the shopkeeper a
s to what his costs are\, and bluffing by the customer as to what she can
afford.\n\nThere is work here too\, going back to Austin\, Searle and Gric
e as well as to economists Crawford and Sobel. There is a nice paper by S
talnaker at MIT which we at CUNY have been in the process of developing an
d we will offer an account of a game theoretic model which resulted from o
ur work.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3355
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3355
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Complexity Aspects of Secure Multiparty Computation
DTSTART;VALUE=DATE-TIME:20140101T090000Z
DTEND;VALUE=DATE-TIME:20140101T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3359@cern.ch
DESCRIPTION:In this talk\, I will survey a collection of results that stud
y qualitative and quantitative complexity of (multi-party) functions\, wit
h respect to how "easy" or "hard" they are for Secure Multi-party Computat
ion.\n\nIn particular\, we formulate a quantitative notion of "cryptograph
ic complexity" of a (multi-party) function\, as the number of "crypto gate
s" needed to securely evaluate the function (amortized over several evalua
tions). Due to recent results\, we can show that up to constants\, this qu
antity does not depend on the specific choice of crypto gate\, as long as
it is "complete."\n\nI shall discuss some connections of cryptographic com
plexity with other notions of complexity of functions\, like circuit compl
exity and communication complexity.\n\nBio: Manoj Prabhakaran is an Associ
ate Professor in the Department of Computer Science at the University of I
llinois\, Urbana-Champaign. Manoj received a Ph.D. in Computer Science fro
m Princeton University in 2005\, and a B.Tech in Computer Science and Engi
neering from the Indian Institute of Technology\, Mumbai\, in 2000. His pr
imary research interest is in theoretical cryptography. His research has b
een supported by an NSF CAREER award and a Beckman faculty fellowship.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3359
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3359
END:VEVENT
BEGIN:VEVENT
SUMMARY:Policy Improvement Algorithm for Zero Sum Two Person Stochastic Ga
mes of Perfect Information
DTSTART;VALUE=DATE-TIME:20140102T103000Z
DTEND;VALUE=DATE-TIME:20140102T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3333@cern.ch
DESCRIPTION:If the data defining a problem and at least one solution to th
e problem lie in the same Archimedean ordered field induced by the data\,
we say that the problem has order field property.\n\nWhen this property ho
lds one may hope to do only finitely many arithmetic operations on the dat
a to arrive at such a solution. For example if we start with rational data
\, the value and a pair of optimal strategies for the players in a zero su
m two person game have rational entries. This was first noticed by Herman
Weyl\, and it was a precursor to solving them via the simplex method. For
bimatrix games while Nash exhibited an equilibrium in mixed strategies\, i
t was Vorobev and Kuhn who checked that the order field property holds for
bimatrix games. Later Lemke and Howson gave the so called linear compleme
ntarity algorithm to locate an equilibrium pair in the same data field. Fo
r three person games\, Nash with Shapley constructed simple counter examp
les for lack of order field property. In general stochastic games fail to
have order field property. \n\nIn our search for finite arithmetic step
algorithms for stochastic games\, perfect information stochastic games sta
nd out for their simplicity of optimal strategies. While we know that they
have order field property\, the key issue is how to adapt the policy imp
rovement algorithm of Markov decision processes to stochastic games. We s
how how to extend this algorithm for zero sum two person stochastic games
of perfect information in discounted as well as Cesaro average payoffs and
show how one avoids cycling.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=3333
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3333
END:VEVENT
BEGIN:VEVENT
SUMMARY:Asymptotic Optimality of Constant-Order Policies for Lost Sales In
ventory Models with Large Lead Times
DTSTART;VALUE=DATE-TIME:20140102T060000Z
DTEND;VALUE=DATE-TIME:20140102T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3367@cern.ch
DESCRIPTION:Lost sales inventory models with large lead times\, which aris
e in many practical settings\, are notoriously difficult to optimize due t
o the curse of dimensionality. In this talk we show that when lead times a
re large\, a very simple constant-order policy\, first studied by Reiman\,
performs nearly optimally. The main insight of our work is that when the
lead time is very large\, such a significant amount of randomness is injec
ted into the system between when an order for more inventory is placed and
when that order is received\, that ``being smart" algorithmically provide
s almost no benefit. Our main proof technique combines a novel coupling fo
r suprema of random walks with arguments from queueing theory (this is joi
nt work with Dmitriy A. Katz-Rogozhnikov\, Yingdong Lu\, Mayank Sharma\, a
nd Mark S. Squillante from IBM T.J. Watson Research Center).\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=3367
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3367
END:VEVENT
BEGIN:VEVENT
SUMMARY:Human-in-the-Loop Embedded Systems: Specification\, Design\, and V
erification
DTSTART;VALUE=DATE-TIME:20140102T083000Z
DTEND;VALUE=DATE-TIME:20140102T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3373@cern.ch
DESCRIPTION:Human interaction is central to many computing systems that re
quire a high level of assurance. Examples of such systems include aircraft
control systems (interacting with a pilot)\, automobiles with self-drivin
g features (interacting with a driver)\, and medical devices (interacting
with a doctor or patient). The correctness of such systems depends not onl
y on the autonomous controller\, but also on the actions of the human cont
roller\, and\, critically\, on the interaction between the two.\n\nIn this
talk\, I will present a formalism for human-in-the-loop embedded systems.
I will then discuss the problem of synthesizing a semi-autonomous control
ler from high-level specifications that expect occasional human interventi
on for correct operation. We present an algorithm for this problem\, and d
emonstrate its operation on problems related to driver assistance in autom
obiles. Finally\, I will present a data-driven approach to probabilistic m
odeling and verification of human driver behavior. Since probabilistic mod
els learned from data can suffer from estimation errors\, we have designed
and applied a new approach for model checking probabilistic models with u
ncertainties. The talk will conclude with a broad survey of research chall
enges in this area.\n\nBio: Sanjit A. Seshia is an Associate Professor in
the Department of Electrical Engineering and Computer Sciences at the Univ
ersity of California\, Berkeley. He received an M.S. and Ph.D. in Computer
Science from Carnegie Mellon University\, and a B.Tech. in Computer Scien
ce and Engineering from the Indian Institute of Technology\, Bombay. His r
esearch interests are in dependable computing and computational logic\, wi
th a current focus on applying automated formal methods to problems in emb
edded systems\, electronic design automation\, computer security\, and pro
gram analysis. His Ph.D. thesis work on the UCLID verifier and decision pr
ocedure helped pioneer the area of satisfiability modulo theories (SMT) an
d SMT-based verification. He is co-author of a widely-used textbook on emb
edded systems. His awards and honors include a Presidential Early Career A
ward for Scientists and Engineers (PECASE) from the White House\, an Alfre
d P. Sloan Research Fellowship\, the School of Computer Science Distinguis
hed Dissertation Award at Carnegie Mellon University\, and the inaugural P
rof. R. Narasimhan Lecture Award from TIFR.\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=3373
LOCATION:Colaba Campus D-406 (D-Block Conference Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3373
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Algorithm for Locating the Nucleolus of Assignment Games
DTSTART;VALUE=DATE-TIME:20140106T103000Z
DTEND;VALUE=DATE-TIME:20140106T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3334@cern.ch
DESCRIPTION:In cooperative game theory\, Shapley value and the nucleolus
are two fundamental solution concepts. They associate a unique imputation
for the players whose coalitional worth $s$ are given a priori.\n\nAss
ignment games are two sided matching games (say between sellers and buyers
) where the coalitional worths between a buyer and seller are specified b
y a nonnegative matrix. (Here\, the rows of the matrix represent sellers
and the columns of the matrix represent buyers). While an optimal assig
nment correspond to best deal\, for real estate agent eyeing on commissio
ns from both sellers and buyers\, he can always initiate an imputation fro
m the core \, via the dual optimal solutions. The algorithm at the end o
f the first iteration reaches the unique corner that favors all sellers. N
ucleolus is invariably a relative interior point of the core and the algo
rithm at each iteration always selects the this most favorable imputation
for all sellers in a shrinking family of imputation polytopes. Here the
shrinking process correspond to the faces of the polytope moving inside at
varying speeds. We can associate with each face of the current polytope\
, the so called unsettled coalitions that clamor for greater satisfaction
for its members. Improvements correspond to redistributions captured throu
gh the notion of squeezing the faces of the polytope by applying differen
t levels of pressure and reducing the dimension of the current subset of t
he imputation set. The algorithm terminates when this squeezing process te
rminates in a single point. The determination of applying such varied pres
sures is achieved through the longest path ending at each vertex of an ass
ociated graph. The algorithm starts with just a set of vertices of a graph
with no arcs to start with. In each intermediate stage\, one introduces
directed arcs corresponding to the equivalence class of unsettled and unha
ppy coalitions. When more and more arcs are introduced\, we reach cycles a
nd the vertices within a cycle are identified as equivalent and are shrunk
to just one representative vertex. This way\, at the end of an iteration\
, the number of vertices of the graph are reduced at least by one. New dir
ected arcs are introduced in the next iteration and this process continues
till just one vertex survives. The algorithm terminates at this stage an
d the associated imputation is the nucleolus. The algorithm exploits the s
pecial lattice structure of the core for assignment games and all computat
ions involve only dealing with the small class of buyer seller pairs.\n\nM
any other subclasses of cooperative games like the connected games and som
e subclass of permutation games have similar algorithms for locating the n
ucleolus. As for locating the Shapley value\, the problems are wide open
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3334
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3334
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction to Polar Codes and Successive Cancellation Decoding
DTSTART;VALUE=DATE-TIME:20140106T083000Z
DTEND;VALUE=DATE-TIME:20140106T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3380@cern.ch
DESCRIPTION:Polar codes have attracted much attention since their introduc
tion by Arikan in 2009 as they are the first type of codes to provably ach
ieve channel capacity. Moreover polar codes have an elegant construction a
s opposed to other modern capacity approaching codes such as LDPC. In this
talk\, I will discuss the phenomenon of polarization of channels which is
fundamental to the functioning of polar codes. Furthermore\, encoding\, S
uccessive Cancellation(SC) decoding\, and hardware architectures for effic
ient SC decoding of polar codes will also be discussed.\n\nBio: Anadi Mish
ra is a PhD candidate at University of Lugano working on automated customi
zation of embedded processors. He obtained BE from NIT-Surat in Electrical
Engineering and MSc. in Embedded System Design from University of Lugano.
He has worked in IT industry for a few years and was Project Assistant at
TIFR before joining MSc. program.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=3380
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3380
END:VEVENT
BEGIN:VEVENT
SUMMARY:From Stability to Differential Privacy
DTSTART;VALUE=DATE-TIME:20140107T060000Z
DTEND;VALUE=DATE-TIME:20140107T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3371@cern.ch
DESCRIPTION:In this talk we establish a connection between some notions of
algorithmic stability and differential privacy. The main thesis is that a
stable algorithm (under certain notions of stability) can be transformed
into a differentially private algorithm with good utility guarantee. In pa
rticular\, we discuss two notions of stability: i) perturbation stability\
, and ii) sub-sampling stability. Based on these notions of stability\, we
provide two generic approaches for the design of differentially private a
lgorithms.\n\nIn the second part of the talk\, we use the generic approach
es designed in the first part to the problem of sparse linear regression i
n high-dimensions. We show that one can design differentially private mode
l (feature) selection algorithms for the above problem. Moreover these alg
orithms have (nearly) optimal sample complexity. We use the celebrated LAS
SO estimator as our basic building block.\n\nBased on joint work with Adam
Smith from Pennsylvania State University.\n\nN.B. As a part of the talk\,
I will give a short tutorial on differential privacy and high-dimensional
statistics. So\, no prior knowledge on either is necessary.\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=3371
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3371
END:VEVENT
BEGIN:VEVENT
SUMMARY:Performance Modeling of Genomics Workflows
DTSTART;VALUE=DATE-TIME:20140110T093000Z
DTEND;VALUE=DATE-TIME:20140110T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3395@cern.ch
DESCRIPTION:Technological advances in sequencing\, mapping\, and analyzing
genomes is proceeding at an extremely rapid pace\, and the resulting expl
osion of genomics data is becoming difficult to manage. Sequencing human g
enomes would quickly add up to hundreds of petabytes of data\, and the dat
a created by analysis of gene interactions multiplies those further. Recen
tly\, the University of Illinois received a grant from the NSF to build a
platform (CompGen) that will allow us to keep ahead of this growth of data
. The CompGen initiative seeks to leverage the strengths of Illinois’ ge
nomic research and that of building large-scale parallel systems\, to deve
lop new technology that allows us to analyze this data more accurately in
a shorter periods of time.\n\nOur approach to this problem is many-fold -
defining new algorithmic techniques\, constructing better metrics for defi
ning accuracy\, extracting parallelism by defining primitives common to th
ese algorithms\, use of accelerators to achieve faster processing\, studyi
ng the effect of architectural changes on the performance of these applica
tions. To unify all of these perspectives and define quantitatively how ea
ch of them effect the system in terms of accuracy and performance\, we con
struct models that describe the applications and system so that\, we can e
xtrapolate the performance metrics to yet to be built systems. We believe
that analyzing such models will give us an understanding of what factors a
re critical in the design of such a system. The problem of designing the C
ompGen system is one of optimization and the information derived from the
model will reduce the complexity of this problem by reducing the design sp
ace of the system.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=3395
LOCATION:Colaba Campus B-333 (DBS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3395
END:VEVENT
BEGIN:VEVENT
SUMMARY:Szemeredi's Regularity Lemma and It's Applications
DTSTART;VALUE=DATE-TIME:20140110T104500Z
DTEND;VALUE=DATE-TIME:20140110T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3399@cern.ch
DESCRIPTION:Szemeredi's Regularity lemma is key result in extremal graph t
heory\, that roughly states the following: For any ϵ>0\, the vertex set o
f a graph G=(V\,E) can be partitioned into k=C(ϵ) (i.e. a constant number
of) parts V1\,...\,Vk of equal size\, such that the induced subgraph on m
ost pairs (Vi\,Vj) is 'ϵ-close' to being a regular bipartite graph. Besid
es many combinatorial applications\, it can also be used to get polynomial
time approximation schemes for various problems on dense graphs. We will
look at a proof of a weakened version of the Regularity Lemma\, and discu
ss a couple of it's applications.\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=3399
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3399
END:VEVENT
BEGIN:VEVENT
SUMMARY:Entropy Power Inequalities: Results and Speculation
DTSTART;VALUE=DATE-TIME:20140113T103000Z
DTEND;VALUE=DATE-TIME:20140113T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3396@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3396
LOCATION:Colaba Campus
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3396
END:VEVENT
BEGIN:VEVENT
SUMMARY:Entropy Power Inequalities: Results and Speculation
DTSTART;VALUE=DATE-TIME:20140113T103000Z
DTEND;VALUE=DATE-TIME:20140113T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3397@cern.ch
DESCRIPTION:Shannon's entropy power inequality characterizes the minimum d
ifferential entropy achievable by the sum of two independent random variab
les with fixed differential entropies.\n\nSince the pioneering work of Sha
nnon\, there has been a steady stream of results over the years\, trying t
o develop parallels to Shannon's entropy power inequality in other scenari
os\, such as for discrete random variables\, and point processes.\n\nWe wi
ll survey this landscape in this talk. We will present old results\, new r
esults\, and share some speculation about how to prove new kinds of entrop
y power inequalities.\n\nhttps://indico.tifr.res.in/indico/conferenceDispl
ay.py?confId=3397
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3397
END:VEVENT
BEGIN:VEVENT
SUMMARY:Connecting Derivative Prices Written on the Same Underlying
DTSTART;VALUE=DATE-TIME:20140114T060000Z
DTEND;VALUE=DATE-TIME:20140114T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3398@cern.ch
DESCRIPTION:Often\, different financial derivatives are subject to the sam
e sources of risk. In such cases\, the prices of these derivatives ought
to be related. In this talk\, we will explore the relationship between a
variance swap\, which is a financial derivative whose payoff depends on th
e entire path of a stock $S$ over a fixed time interval $[0\,T]$\, and a E
uropean contract\, which is a financial derivative whose payoff depends on
ly on the value of $S$ at the terminal time $T$. We will prove that\, whe
n a stock is modeled as a time-changed Markov process\, the variance swap
has the same value as a European contract whose payoff function is the sol
ution of an integro-differential equation (which we will solve). The sign
ificance of this result is that the path-dependent variance swap contract
can be priced relative to liquidly traded (and efficiently priced) path-in
dependent European options in a semi-nonparametric fashion.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=3398
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3398
END:VEVENT
BEGIN:VEVENT
SUMMARY:Interior Point Methods (as seen by Renegar)
DTSTART;VALUE=DATE-TIME:20140117T090000Z
DTEND;VALUE=DATE-TIME:20140117T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3407@cern.ch
DESCRIPTION:Interior point methods or IPMs are a class of iterative algori
thms in optimization that follow an interior path (unlike the simplex meth
od which follows the boundary). Von Neumann proposed an IPM for linear pro
gramming (LP) way back in 1948. Later IPMs were generalized for convex opt
imization\, before their reappearance as fast LP solvers with Karmarkar's
algorithm. Subsequently\, Nesterov-Nemirovskii proved fast convergence of
IPMs for convex optimization using "self-concordant barrier" functions.\n\
nIn his book\, Renegar revisits this very creative but mysterious work of
Nesterov-Nemirovskii\, and tries to demystify it from a mathematical viewp
oint. I will try to explain this approach in my talk. The talk will be sel
f-contained and assume only basic knowledge of linear algebra and vector c
alculus.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3407
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3407
END:VEVENT
BEGIN:VEVENT
SUMMARY:When Heavy-tailed and Light-tailed Flows Compete
DTSTART;VALUE=DATE-TIME:20140120T060000Z
DTEND;VALUE=DATE-TIME:20140120T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3360@cern.ch
DESCRIPTION:This talk focuses on the design and analysis of scheduling pol
icies for multi-class queues\, such as those found in wireless networks an
d high-speed switches. In this context\, we study the response time tail
under generalized max-weight policies in settings where the traffic flows
are highly asymmetric. Specifically\, we consider a setting where a bursty
flow\, modeled using heavy-tailed statistics\, competes with a more benig
n\, light-tailed flow. In this setting\, we prove that classical max-weigh
t scheduling\, which is known to be throughput optimal\, results in the li
ght-tailed flow having heavy-tailed response times. However\, we show that
via a careful design of inter-queue scheduling policy (from the class of
generalized max-weight policies) and intra-queue scheduling policies\, it
is possible to maintain throughput optimality\, and guarantee light-tailed
delays for the light-tailed flow\, without affecting the response time ta
il for the heavy-tailed flow.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=3360
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3360
END:VEVENT
BEGIN:VEVENT
SUMMARY:Exact Simulation of Multidimensional Reflected Diffusions
DTSTART;VALUE=DATE-TIME:20140121T103000Z
DTEND;VALUE=DATE-TIME:20140121T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3408@cern.ch
DESCRIPTION:A number of algorithms have been produced recently to simulate
exactly from diffusions (typically one dimensional). A key idea in all of
these algorithms is to propose from a process which is absolutely continu
ous with respect to the target process. The likelihood ratio can often be
suitably localized and the so-called acceptance step can be executed using
a thinning procedure. While similar strategies can be used for one dimens
ional reflected diffusions\, multidimensional reflected diffusions\, call
for a new methodology. This is mainly because there is no simulatable proc
ess that can be used as a proposal. In this talk we explain this new metho
dology (this talk is based on joint work with Chris Dolan and Karthyek Mur
thy).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=340
8
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3408
END:VEVENT
BEGIN:VEVENT
SUMMARY:Switching in Boolean Circuits and Modelling Cognition Through Neur
oids
DTSTART;VALUE=DATE-TIME:20140121T083000Z
DTEND;VALUE=DATE-TIME:20140121T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3411@cern.ch
DESCRIPTION:In computational complexity theory\, Håstad's switching lemma
is a vital analytical tool for proving lower bounds on the size of consta
nt-depth Boolean circuits. Using the switching lemma\, Håstad showed in 1
986 that Boolean circuits of depth $k$\, in which only AND\, OR\, and NOT
gates are allowed\, require size $\\exp\\left(\\Omega\\left(n^{\\frac{1}{k
-1}}\\right)\\right)$ for computing the PARITY function on $n$ variables.
In essence\, the switching lemma says that\, given an arbitrary formula in
disjunctive normal form\, if we set some fraction of the variables random
ly\, then with high probability\, the restricted function can be computed
by a decision tree of small depth. In 2012\, Impagliazzo\, Matthews and Pa
turi formulated an extended switching lemma which says that\, given a sequ
ence of formulas in conjunctive and/or disjunctive normal form on the same
set of variables\, if all of them are hit by the same random restriction\
, then it is exponentially unlikely that there is a large subset of formul
as where each formula contributes a large number of variables to their joi
nt decision tree. In the first part of my talk\, we will discuss the exten
ded switching lemma and its proof with the help of an illustrative example
.\n\nIn the second part of the talk\, we will critique the neuroidal model
for cognition proposed by Valiant. We will start off by discussing the mo
tivation behind Valiant's work. We will then explore the physiology of the
brain and some insights from cognitive psychology that aided Valiant's fo
rmulation of the neuroidal model. Next\, we will describe the actual model
in extensive detail and present the algorithm put forward by Valiant for
implementing unsupervised memorization within his model as a case study. F
inally\, we will conclude by discussing the relevance of the neuroidal mod
el to the current attempts by researchers to build cognitive computing sys
tems.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=341
1
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3411
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Communication Complexity of Secure Computation
DTSTART;VALUE=DATE-TIME:20140124T100000Z
DTEND;VALUE=DATE-TIME:20140124T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3412@cern.ch
DESCRIPTION:In Secure multiparty computation (MPC)\, mutually distrustful
parties each having a private input\, perform a joint computation without
revealing their inputs to each other. Information theoretically secure mul
ti-party computation has been a central primitive of modern cryptography.
However\, relatively little is known about the communication complexity of
this primitive.\n\nIn this talk\, first I will give some background in se
cure MPC and information theory. We consider the problem where three parti
es are involved\, two of them (Alice and Bob) have inputs and third one (C
harlie) wants to compute the output. We insist that Charlie does not learn
anything about the other parties' inputs from the protocol other than his
output and whatever it reveals about the inputs. Also\, Alice and Bob do
not learn anything about the other parties' input/output from the protocol
. We develop information theoretic tools to prove lower bounds on the comm
unication complexity of secure computation in our setting and obtain tight
bounds for secure computation protocols for various interesting functions
. In particular\, we show that for certain functions\, there are “commun
ication-ideal” protocols\, which achieve the minimum communication simul
taneously on all links in the network.\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=3412
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3412
END:VEVENT
BEGIN:VEVENT
SUMMARY:Preserving Terminal Distances Using Minors
DTSTART;VALUE=DATE-TIME:20140131T083000Z
DTEND;VALUE=DATE-TIME:20140131T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3442@cern.ch
DESCRIPTION:We ask the following question: given a network with a large nu
mber of nodes\, edges with nonnegative costs on them and a small set of sp
ecial nodes called terminals\, by how much can we compress the graph such
that the shortest distance between any pair of terminals remains the same?
We are allowed to perform a sequence of the following operations:\n\n(1)
Delete an edge (this reduces the number of edges by 1)\n\n(2) Contract (or
shrink) an edge into a single node (this reduces the number of edges and
number of nodes by 1)\n\n(3) Change edge cost arbitrarily\, as long as it
remains nonnegative\n\nFormally stated\, for any family $\\mathcal{F}$ of
undirected graphs\, what is the smallest $f(k\,\\mathcal{F})$ such that ev
ery graph $G(V\,E) \\in\n\\mathcal{F}$ with nonnegative weights on the edg
es and a set of terminal nodes $R \\subseteq V$ having $|R|=k$ admits a di
stance preserving minor with at most $f(k\,\\mathcal{F})$ nodes? (Observe
that $f$ does not depend on the total number of nodes in the original grap
h).\n\nIn this talk\, we will be defining the terms minor and distance pre
serving minor with examples. If $\\mathcal{G}$ is the family of all undire
cted\ngraphs\, $\\mathcal{P}$ is the family of all planar graphs and $\\ma
thcal{T}$ is the family of all trees\, we will show that:\n\n$f(k\,\\mathc
al{G}) \\leq k^{4}$\n$f(k\,\\mathcal{P}) \\geq c \\cdot k^{2}$ (for some c
onstant $c$)\n$f(k\,\\mathcal{T})=2k-2$\n\nThis is due to recent work by R
obert Krauthgamer and Tamar Zondiner.\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=3442
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3442
END:VEVENT
BEGIN:VEVENT
SUMMARY:Online Algorithms for Node-weighted Network Design
DTSTART;VALUE=DATE-TIME:20140203T101500Z
DTEND;VALUE=DATE-TIME:20140203T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3436@cern.ch
DESCRIPTION:In recent years\, an online adaptation of the classical primal
-dual paradigm has been successfully used to obtain new online algorithms
for node-weighted network design problems\, and simplify existing ones for
their edge-weighted counterparts. In this talk\, I will give an outline o
f this emerging toolbox using three fundamental problems in this category
for illustration: the Steiner tree problem (Naor-P.-Singh\, 2011)\, the St
einer forest problem (Hajiaghayi-Liaghat- P.\, 2013)\, and their respectiv
e prize-collecting variants (Hajiaghayi-Liaghat-P.\, 2014).\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=3436
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3436
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Three Parameters of (non-)Convexity of Sets in Euclidean Space
DTSTART;VALUE=DATE-TIME:20140204T060000Z
DTEND;VALUE=DATE-TIME:20140204T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3447@cern.ch
DESCRIPTION:We consider three different parameters measuring (non-)convexi
ty of subsets of the plane or\, more generally\, of an Euclidean space. I
n particular\, we are interested in mutual relations of these parameters.\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3447
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3447
END:VEVENT
BEGIN:VEVENT
SUMMARY:Resource Allocation Games and Pricing in Complex Priority Systems
DTSTART;VALUE=DATE-TIME:20140205T103000Z
DTEND;VALUE=DATE-TIME:20140205T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3448@cern.ch
DESCRIPTION:Differentiated service is typically provided to customers by o
ffering different grades of service at diﬀerent prices. We ﬁrst consid
er a simple system with heterogeneous customers that are type indexed by $
v$. In this system\, a customer chooses one of a continuum of service grad
es that the server provides. Higher service grades are priced higher by th
e server. Different customers value paying this price differently\, which
is described by an arbitrary weighting function that is decreasing in $v$.
We show that at Nash equilibrium\, the revenue to the service provider is
independent of how the server chooses to price different service grades.\
n\nNext we consider a complex system where a customer has to choose two pr
iority parameters instead of one in simple systems. The service provider u
ses complex scheduling to provide a grade of service that is a function of
the two parameters\; higher values of the parameters provide better servi
ce grades. The two priority parameters have different cost functions that
are increasing in the value of the parameter. Further\, the two parameters
are weighted differently by different types of customers with one of them
having weights increasing while the other decreasing in v. We analyze the
Nash equilibrium when customers are selﬁsh in such a complex system and
characterize the QoS for each customer type at Nash equilibrium. We then
consider pricing the services to maximize the revenue for the service prov
ider in an example system. Examples from diverse domains will show that th
e results are very general.\n\nhttps://indico.tifr.res.in/indico/conferenc
eDisplay.py?confId=3448
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3448
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Hypothesis Testing Approach for Proving Inner Bounds for Some In
formation Theoretic Problems
DTSTART;VALUE=DATE-TIME:20140207T090000Z
DTEND;VALUE=DATE-TIME:20140207T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3456@cern.ch
DESCRIPTION:Most of the achievability proofs in Information theory are bas
ed on the concept of typicality. However\, this typicality technique does
not seem to be good enough to give achievability bounds in the most genera
l settings\, i.e. in the one-shot or non-iid case. In this talk we will di
scuss a hypothesis testing technique for proving inner bounds for some inf
ormation theoretic problems. In particular\, we will discuss about using h
ypothesis testing technique to prove achievable rates for point to point a
nd multiple access channel. If time permits we will also discuss the exten
sion of this technique in the quantum case\, i.e.\, when the classical mes
sage is sent over the quantum channel.\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=3456
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3456
END:VEVENT
BEGIN:VEVENT
SUMMARY:High-Dimensional Inverse Problems in Seismology
DTSTART;VALUE=DATE-TIME:20140210T103000Z
DTEND;VALUE=DATE-TIME:20140210T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3440@cern.ch
DESCRIPTION:The inference of the interior properties of the Sun and Earth
using surface measurements of the seismic wavefield are active areas. I wi
ll discuss the construction of accurate forward models for seismic wavefie
lds and the solution of the inverse problem using PDE-constrained optimiza
tion. These techniques are useful when attempting to invert for the struct
ure of strong 3-D heterogeneities such as sunspots and Earth's crustal str
ucture/exploration.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=3440
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3440
END:VEVENT
BEGIN:VEVENT
SUMMARY:Bridging the Gap Between Classical Logic Based Formalisms and Logi
c Programs
DTSTART;VALUE=DATE-TIME:20140211T060000Z
DTEND;VALUE=DATE-TIME:20140211T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3454@cern.ch
DESCRIPTION:Different logic-based knowledge representation formalisms have
different limitations either with respect to expressivity or with respect
to computational efficiency. First-order logic\, which is the basis of De
scription Logics (DLs)\, is not suitable for defeasible reasoning due to i
ts monotonic nature. The nonmonotonic formalisms that extend first-order l
ogic\, such as circumscription and default logic\, are expressive but lack
efficient implementations. The nonmonotonic formalisms that are based on
the declarative logic programming approach\, such as Answer Set Programmin
g (ASP)\, have efficient implementations but are not expressive enough for
representing and reasoning with open domains.\n\nWe use the first-order s
table model semantics\, which extends both first-order logic and ASP\, to
relate circumscription to ASP\, and to integrate DLs and ASP\, thereby par
tially overcoming the limitations of the formalisms. By exploiting the rel
ationship between circumscription and ASP\, we show how well-known action
formalisms\, such as the situation calculus\, the event calculus\, and Tem
poral Action Logics\, can be reformulated in ASP. We show that these refor
mulations have certain advantages with respect to the generality of the re
asoning tasks that can be handled and with respect to the computational ef
ficiency. Our integration of DLs and ASP enables us to perform nonmonotoni
c reasoning with DL knowledge bases and provides a framework for integrati
ng rules and ontologies for the semantic web. Observing the need to integr
ate action theories and ontologies\, we use the above results to reformula
te the problem of integrating action theories and ontologies as a problem
of integrating rules and ontologies\, thus enabling us to use the computat
ional tools developed in the context of the latter for the former.\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3454
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3454
END:VEVENT
BEGIN:VEVENT
SUMMARY:Preferential Attachment Random Graphs With General Weight Function
DTSTART;VALUE=DATE-TIME:20140213T090000Z
DTEND;VALUE=DATE-TIME:20140213T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3457@cern.ch
DESCRIPTION:We will start with the classical Albert - Barabasi model of pr
eferential attachment random graphs. We shall prove rigorously using an em
bedding in Markov branching processes the power law growth of degrees and
the power law decay of the limiting empirical distribution of the degree.
Next we will consider the case of general weight function and show the dif
ferent behaviours for the superlinear\, linear and sublinear cases.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3457
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3457
END:VEVENT
BEGIN:VEVENT
SUMMARY:Perfect Sampling for Doeblin Chains
DTSTART;VALUE=DATE-TIME:20140213T053000Z
DTEND;VALUE=DATE-TIME:20140213T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3458@cern.ch
DESCRIPTION:Propp and Wilson showed how to generate a Markov chain which i
n a finite number of steps gives a sample from the stationary distribution
supported by a countable set. In this talk we show that this is feasible
if the chain satisfies the Doeblin recurrence condition.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=3458
LOCATION:Colaba Campus D-406 (D-Block Conference Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3458
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Past and Future of Brownian Motion
DTSTART;VALUE=DATE-TIME:20140214T053000Z
DTEND;VALUE=DATE-TIME:20140214T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3462@cern.ch
DESCRIPTION:Let B be standard Brownian motion. Fix an interval (a\,b). Con
dition on B(t) to be in (a\,b). Look at B(u) for u<.=t and B(u) u>.=t. We
show that this converges weakly to a proper probability measure on C(R).\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3462
LOCATION:Colaba Campus AG-77
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3462
END:VEVENT
BEGIN:VEVENT
SUMMARY:When is a Formula Invariant?
DTSTART;VALUE=DATE-TIME:20140221T103000Z
DTEND;VALUE=DATE-TIME:20140221T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3481@cern.ch
DESCRIPTION:Program invariants play an important useful role in understand
ing programs as well as verifying properties about them. There are many wa
ys to obtain program invariants--automatically\, semi-automatically or the
y could be provided by a programmer as a part of a specification of the pr
ogram or as annotations. The problem of determining whether a proposed for
mula is an invariant is undecidable. In this talk\, methods are proposed t
o determine when in certain cases\, formulas expressed in various logical
theories can be determined to be invariant. If a formula is indeed an inva
riant\, techniques for strengthening the formula to generate inductive inv
ariants are explored.\n\nhttps://indico.tifr.res.in/indico/conferenceDispl
ay.py?confId=3481
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3481
END:VEVENT
BEGIN:VEVENT
SUMMARY:Collisions Between Random Walks
DTSTART;VALUE=DATE-TIME:20140224T103000Z
DTEND;VALUE=DATE-TIME:20140224T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3482@cern.ch
DESCRIPTION:An infinite graph may have the property that two independent r
andom walks on the graph never meet each other\, although each of them vi
sits every vertex of the graph infinitely often. I shall describe some old
results joint with Yuval Peres and later results of Barlow\, Peres and So
usi. Finally I shall report some ongoing recent investigations in this dir
ection.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3
482
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3482
END:VEVENT
BEGIN:VEVENT
SUMMARY:Zeros of Random Polynomials
DTSTART;VALUE=DATE-TIME:20140225T090000Z
DTEND;VALUE=DATE-TIME:20140225T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3483@cern.ch
DESCRIPTION:We give an overview of zeros of random polynomials. Starting w
ith results dating back to 1930s (Mark Kac\, Offord etc)\, we intend to co
ver some of the many recent developments (due to various authors in the pa
st ten years). These will include central limit theorems\, large deviation
probabilities\, matching theorems etc.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=3483
LOCATION:Colaba Campus AG-66 (Leecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3483
END:VEVENT
BEGIN:VEVENT
SUMMARY:Understanding Global Properties of Data Sets from Local Observatio
ns
DTSTART;VALUE=DATE-TIME:20140225T043000Z
DTEND;VALUE=DATE-TIME:20140225T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3486@cern.ch
DESCRIPTION:Suppose we are given a list of numbers and we wish to determin
e whether it is sorted in increasing order. That problem obviously require
s reading the entire list. However\, it turns out that if we know in advan
ce that our list is either sorted or far from sorted\, we can perform the
test by examining only a small portion of the list. This is an example of
a global property of a data set that we can understand by making a few loc
al observations.\nAs data of all types gets easier to obtain and cheaper t
o store\, data sets are becoming increasingly large. Consequently\, there
is a need to perform computational tasks on massive data sets. What useful
computations can be performed on a data set when reading all of it is pro
hibitively expensive? This question\, fundamental to several fields\, is a
t the heart of a research area\, called Sublinear Algorithms\, that has pr
ovided important insights into fast approximate computation.\n\nIn this ta
lk\, we will give a few examples of specific problems that can be solved w
hile making only local observations\, starting with the sorting example an
d moving on to simple analysis of images\, comparing and compressing docum
ents\, and understanding properties of functions. We will also present an
application of these ideas to an area (namely\, data privacy)\, where extr
eme efficiency is not a requirement per se\, but helps to guarantee other
properties.\n\nSpeaker Bio: Sofya Raskhodnikova is an associate professor
of Computer Science and Engineering at Penn State. She received her Ph.D.
from MIT. Prior to joining Penn State in 2007\, she was a postdoctoral fel
low at the Hebrew University of Jerusalem and the Weizmann Institute of Sc
ience\, and a visiting scholar at the Institute for Pure and Applied Mathe
matics at UCLA. Currently\, she is on sabbatical at Boston University.\n\n
Dr. Raskhodnikova works in the areas of randomized and approximation algor
ithms. Her main interest is the design and analysis of sublinear-time algo
rithms for combinatorial problems. Sublinear algorithms produce approximat
e answers after examining only a tiny portion of the data. Such algorithms
are important for dealing with massive data sets.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=3486
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3486
END:VEVENT
BEGIN:VEVENT
SUMMARY:Private Analysis of Graphs
DTSTART;VALUE=DATE-TIME:20140225T103000Z
DTEND;VALUE=DATE-TIME:20140225T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3487@cern.ch
DESCRIPTION:We discuss algorithms for the private analysis of network data
. Such algorithms work on data sets that contain sensitive relationship in
formation (for example\, romantic ties). Their goal is to compute approxim
ations to global statistics of the graph while protecting information spec
ific to individuals. Our algorithms satisfy a rigorous notion of privacy\,
called node differential privacy. Intuitively\, it means that an algorith
m's output distribution does not change significantly when a node and all
its adjacent edges are removed from a graph.\n\nA key component of our wor
k is the design of efficiently computable Lipschitz extensions of commonly
computed graph statistics. Given a graph statistic f\, we seek to design
a new function g that is efficiently computable and "robust" to the additi
on or removal of vertices\, yet agrees with f on as large a set of graphs
as possible. Our techniques are based on combinatorial analysis\, network
flow\, and linear and convex programming.\n\nBased on joint work with Shiv
a Kasiviswanathan\, Kobbi Nissim and Sofya Raskhodnikova.\n\nBio: Adam Smi
th is an associate professor in the Department of Computer Science and Eng
ineering at Penn State\; currently\, he is on sabbatical at Boston Univers
ity. His research interests lie in cryptography\, privacy and their connec
tions to information theory\, quantum computing and statistics. He receive
d his Ph.D. from MIT in 2004 and was subsequently a visiting scholar at th
e Weizmann Institute of Science and UCLA.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=3487
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3487
END:VEVENT
BEGIN:VEVENT
SUMMARY:Coverage and Capacity Analysis of mmWave Cellular Systems
DTSTART;VALUE=DATE-TIME:20140226T103000Z
DTEND;VALUE=DATE-TIME:20140226T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3494@cern.ch
DESCRIPTION:Millimeter wave (mmWave) spectrum may be the solution to the s
pectrum gridlock in cellular systems. mmWave systems overcome potentially
high pathloss by using large antenna arrays at both the transmitter and re
ceiver\, to provide enough beamforming gain to reverse\, if not benefit fr
om\, the effects of the higher carrier. In this talk\, we introduce the co
ncept of mmWave cellular systems. Then we examine the system-level perform
ance of mmWave cellular systems with a special focus on coverage and capac
ity. This talk presents an analysis of mmWave cellular systems using the m
athematical framework of stochastic geometry\, which has been used to anal
yze microwave cellular and ad-hoc networks. The analysis incorporates mmWa
ve's key differentiating factors such as the limited scattering nature of
mmWave channels\, and the use of RF beamforming strategies (also known as
beam steering) to provide highly directional transmission with limited har
dware complexity. To model mmWave signals' increased susceptibility to sig
nal blockage (shadowing) in urban environments\, an exciting new tool is l
everaged known as random shape theory to model blockages due to buildings.
The results show that\, in general\, coverage in mmWave systems can rival
or even exceed coverage in microwave systems assuming that the link margi
ns promised by existing mmWave system designs are in fact achieved. This c
omparable coverage translates into a superior average rate performance for
mmWave systems as a result of the larger bandwidth available for transmis
sion.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=349
4
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3494
END:VEVENT
BEGIN:VEVENT
SUMMARY:Determinantal Point Processes
DTSTART;VALUE=DATE-TIME:20140228T090000Z
DTEND;VALUE=DATE-TIME:20140228T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3484@cern.ch
DESCRIPTION:Determinantal point processes form an important class of point
processes with negative correlations ("repulsion"). We give an overview s
tarting from definitions. We intend to present a few general results and s
everal examples\, in particular random spanning trees.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=3484
LOCATION:Colaba Campus AG-66 (Leecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3484
END:VEVENT
BEGIN:VEVENT
SUMMARY:Low-rank Matrix and Tensor Recovery: Theory and Algorithms
DTSTART;VALUE=DATE-TIME:20140303T103000Z
DTEND;VALUE=DATE-TIME:20140303T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3501@cern.ch
DESCRIPTION:Recovering a low-rank matrix or tensor from incomplete or corr
upted observations is a recurring problem in signal processing and machine
learning. To exploit the structure of data that is intrinsically more tha
n three-dimensional\, convex models such low-rank completion and robust pr
incipal component analysis (RPCA) for matrices have been extended to tenso
rs. In this talk\, we rigorously establish recovery guarantees for both te
nsor completion and tensor RPCA. We demonstrate that using the most popula
r convex relaxation for the tensor Tucker rank can be substantially subopt
imal in terms of the number of observations needed for exact recovery. We
introduce a very simple\, new convex relaxation which is shown be much bet
ter\, both theoretically and empirically. Moreover\, we propose algorithms
to solve both low-rank matrix and tensor recovery models based on the Alt
ernating Direction Augmented Lagrangian (ADAL)\, Frank-Wolfe and prox-grad
ient methods. Finally\, we empirically investigate the recoverability prop
erties of these convex models\, and the computational performance of our a
lgorithms on both simulated and real data (This is joint work with Cun Mu\
, Bo Huang and Tony Qin (IEOR PhD students at Columbia University) and Joh
n Wright (E.E. faculty member at Columbia University)).\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=3501
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3501
END:VEVENT
BEGIN:VEVENT
SUMMARY:Diffusion Approximation to Adaptive MCMC
DTSTART;VALUE=DATE-TIME:20140304T060000Z
DTEND;VALUE=DATE-TIME:20140304T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3469@cern.ch
DESCRIPTION:Adaptive MCMC algorithms are designed to tune the chain proper
ly to decrease the time to convergence to stationary. However since the tr
ansition kernel is changed at each step of iteration proper ergodicity con
ditions has to be ensured. Here we define a discrete time Adaptive MCMC al
gorithm and study its convergence properties. We apply the diffusion appro
ximation method to a discrete time Adaptive MCMC procedure to obtain the l
imiting stochastic differential equation governing the dynamics of the ada
ptation parameter $\\theta$ and the state space variable $X$. The solution
to the coupled equation will give the stationary distribution of the chai
n. Comparison of rates of convergence between the standard and the adaptiv
e method will be of interest.\n\nhttps://indico.tifr.res.in/indico/confere
nceDisplay.py?confId=3469
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3469
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sparsest Cut in Bounded Treewidth Graphs
DTSTART;VALUE=DATE-TIME:20140305T103000Z
DTEND;VALUE=DATE-TIME:20140305T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3503@cern.ch
DESCRIPTION:We consider approximation algorithms for the sparsest cut grap
h partitioning problem. Here\, given graphs G with demand pairs $\\{s_i\,t
_i\\}$\, the goal is to separate G into two parts to minimize the ratio of
the number of edges cut to the number of demand pairs separated.\n\nIn th
is talk\, I will sketch:\n- the best hardness-of-approximation (based on P
!=NP) currently known for this problem\,\n- and a 2-approximation algorith
m with running time $n^{O(k)}$\, where $k$ is the treewidth of the underly
ing graph G. Our algorithm rounds a Sherali-Adams LP relaxation.\n\nThis p
ositive result can be complemented by (a) an integrality gap of a factor o
f 2 for the Sherali-Adams hierarchy even after polynomially\nmany rounds\,
and (b) an unique-games hardness of a factor of 2. Time permitting\, I wi
ll give a high-level intuition of how the NP-hardness\ncan be extended to
prove these matching results.\n\nI will try to keep the talk as self-conta
ined as possible (this is joint work with Kunal Talwar (Microsoft Research
SVC) and David Witmer (CMU)\, and appeared at the STOC 2013 conference).\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3503
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3503
END:VEVENT
BEGIN:VEVENT
SUMMARY:How to Run your Chores\, and Get to Dinner on Time
DTSTART;VALUE=DATE-TIME:20140306T090000Z
DTEND;VALUE=DATE-TIME:20140306T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3504@cern.ch
DESCRIPTION:In the orienteering problem\, we are given a metric space (the
distances are supposed to represent travel times between the locations)\,
a start vertex ("home") and a deadline B\, and want to visit as many poin
ts as possible using a tour of length at most B. We know constant-factor a
pproximation algorithms for this problem. However\, suppose it is not enou
gh for us to visit the nodes: upon reaching a location\, we also have to w
ait for some time at each location before we can get the reward. Each such
waiting time is drawn from a known probability distribution. What can we
do then? In this talk\, we will discuss adaptive and non-adaptive approxim
ation algorithms for this stochastic orienteering problem (his is based on
work with Ravi Krishnaswamy\, Viswanath Nagarajan\, and R. Ravi).\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3504
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3504
END:VEVENT
BEGIN:VEVENT
SUMMARY:First Order Logics Extended With Counting Quantifiers
DTSTART;VALUE=DATE-TIME:20140312T043000Z
DTEND;VALUE=DATE-TIME:20140312T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3517@cern.ch
DESCRIPTION:In this talk\, we shall look at logics over words. It is known
that first order logic (FO) with an order relation\, \n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=3517
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3517
END:VEVENT
BEGIN:VEVENT
SUMMARY:A PMU Scheduling Scheme for Transmission of Synchrophasor Data in
Electric Power Systems
DTSTART;VALUE=DATE-TIME:20140313T053000Z
DTEND;VALUE=DATE-TIME:20140313T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3498@cern.ch
DESCRIPTION:With the proposition to install a large number of phasor measu
rement units (PMUs) in the future electric grid\, it is essential to provi
de robust communications infrastructure for phasor data across the network
. We make progress in this direction by devising a simple time division mu
ltiplexing scheme for transmission of phasor data from the PMUs to a centr
al server: time is divided into frames and a particular set of PMUs within
a given time frame take turns to transmit to the control center. The main
contribution of this work is a scheduling policy based on which the PMUs
take turns during a time frame. Let N be the optimal number of PMUs obtain
ed by solving the PMU placement problem. The scheduling scheme is develope
d for two scenarios: (a) topology-based PMU placement\, when N is time-inv
ariant and (b) when N is a function of time\, due to the dynamic nature of
the system states. Unlike strategies devised for conventional communicati
ons\, the scheduling scheme presented here is intended for the power netwo
rk\, since it is governed by the measure of electrical connectedness betwe
en various components in the grid. The problem is approached from the pers
pective of complex networks\, which promotes the electrical structure of t
he grid over its topological structure. Relative merits of the proposed sc
heduling scheme over conventional techniques employed in practice are demo
nstrated via simulations. \n\n \nBio: Nagananda was born and raised in Ban
galore\, India. He got his B.Tech in telecommunications engineering from V
isveswaraiah Technological University in 2003. He got his M.S. in electric
al engineering from Oregon Health and Science University\, Portland\, Oreg
on in 2007\, and his PhD in electrical engineering from Lehigh University
in 2013. His was affiliated with the ECE department at the Indian Institut
e of Science in various capacities. He is also an adjunct associate profes
sor at the People's Education Society Institute of Technology\, Bangalore.
His main areas of interest are in information theory and communications a
imed at solving infrastructure-related problems in the electric power grid
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3498
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3498
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Group of Symmetries of the Rubik's Cube
DTSTART;VALUE=DATE-TIME:20140314T090000Z
DTEND;VALUE=DATE-TIME:20140314T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3529@cern.ch
DESCRIPTION:In this talk\, we will prove that the group of symmetries of a
standard (3x3x3) Rubik's cube is isomorphic to (\\mathbb{Z}_37 \\times \\
mathbb{Z}_2^{11}) \\rtimes ((A_8 \\times A_{12}) \\rtimes Z_2). Due to lim
ited time\, some proofs may not be completely rigorous. We will also try t
o understand how the above structure suggests a natural commutator based a
pproach to solving the Rubik's cube (and also how it generalizes to higher
order cubes).\n\nThe talk will assume a very basic group theory backgroun
d\, namely knowledge of cyclic groups and direct products. We will define
semidirect products and alternating groups through the course of the talk.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3529
LOCATION:Colaba Campus
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3529
END:VEVENT
BEGIN:VEVENT
SUMMARY:Parameterized Algorithms for Minimum Vertex Cover Problem
DTSTART;VALUE=DATE-TIME:20140321T090000Z
DTEND;VALUE=DATE-TIME:20140321T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3549@cern.ch
DESCRIPTION:An optimization problem is said to be fixed parameter tractabl
e if there is some parameter k independent of the input size n\, such that
the problem can be solved in O(f(k).n^c) time\, where c is a constant\, a
nd f is an arbitrary function depending only on k. So\, for instances of N
P-Hard problems where k is considerably small despite large sizes of n\, s
uch parameterized algorithms run in practically affordable time. In this t
alk we will go through some parameterized algorithms for the minimum verte
x cover problem of simple graphs.\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=3549
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3549
END:VEVENT
BEGIN:VEVENT
SUMMARY:Bernoulli Factory
DTSTART;VALUE=DATE-TIME:20140328T090000Z
DTEND;VALUE=DATE-TIME:20140328T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3567@cern.ch
DESCRIPTION:Necessary and sufficient conditions on a function $f(p)$ are g
iven for the existence of a simulation procedure to simulate a Bernoulli r
andom variable with success probability $f(p)$ from independent Bernoulli
random variables with success probability $p$\, with $p$ being constrained
to lie in a subset of $[0\,1]$ but otherwise unknown.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=3567
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3567
END:VEVENT
BEGIN:VEVENT
SUMMARY:Central Limit Theorems for Some Random Simplicial Complexes
DTSTART;VALUE=DATE-TIME:20140331T090000Z
DTEND;VALUE=DATE-TIME:20140331T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3563@cern.ch
DESCRIPTION:In this talk\, we shall state central limit theorems for many
local and global functionals of simplicial complexes built on various rand
om point processes. In the first part of the talk we will consider simplic
ial counts in Cech and Voronoi simplicial complexes for long-range depende
nt point processes such as zeros of Gaussian analytic functions and determ
inantal point processes. These functionals serve as a good illustration of
our general central limit theorems for local functionals of the above poi
nt processes.\n\nIn the second part\, we shall restrict ourselves to the u
biquitous Poisson point process but look at a very global functional calle
d the Betti number. We shall show various stabilizing properties of the Be
tti numbers of the random Cech complex to leverage recent results on stabi
lizing functionals of Poisson point processes.\n\nhttps://indico.tifr.res.
in/indico/conferenceDisplay.py?confId=3563
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3563
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hierarchical Coded Caching: Divide and Conquer
DTSTART;VALUE=DATE-TIME:20140404T053000Z
DTEND;VALUE=DATE-TIME:20140404T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3542@cern.ch
DESCRIPTION:Caching of popular content during off-peak hours is a strategy
to reduce network loads during peak hours. Recent work has shown signific
ant benefits of designing such caching strategies not only to deliver part
of the content locally\, but also to provide coded multicasting opportuni
ties even among users with different demands. Exploiting both of these gai
ns has been shown to be approximately optimal for caching systems with a s
ingle layer of caches.\n\nMotivated by practical scenarios\, we consider a
hierarchical content delivery network with two layers of caches. We propo
se a new caching scheme that combines two basic approaches. The first appr
oach uses a decode and forward strategy to provide coded multicasting oppo
rtunities within each layer\; the second approach uses strategic forwardin
g without decoding to provide coded multicasting opportunities across the
two layers. By striking the right balance between these two approaches\, w
e show that the proposed scheme achieves the optimal communication rates t
o within a constant multiplicative and additive gap. We further show that
there is no tension between the rates in each of the two layers up to the
aforementioned gap. Thus\, both layers can simultaneously operate at appro
ximately the minimum rate.\n\nBio: Nikhil Karamchandani is a postdoctoral
researcher in the Department of Electrical Engineering at the University o
f California Los Angeles. He received the B.Tech degree from the Departmen
t of Electrical Engineering at the Indian Institute of Technology Bombay i
n 2005\, and the M.S. and Ph.D. degrees from the Department of Electrical
and Computer Engineering at the University of California at San Diego in 2
007 and 2011 respectively. His research interests are in networks\, commun
ications\, and information theory. Mr. Karamchandani received the Californ
ia Institute for Telecommunications and Information Technology (CalIT2) fe
llowship in 2005.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.p
y?confId=3542
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3542
END:VEVENT
BEGIN:VEVENT
SUMMARY:Safe Execution of (Untrusted) Applications Downloaded From the Int
ernet
DTSTART;VALUE=DATE-TIME:20140404T090000Z
DTEND;VALUE=DATE-TIME:20140404T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3569@cern.ch
DESCRIPTION:From a security perspective\, in the current age where malware
(malicious software) distribution is rampant\, it is dangerous to execute
programs downloaded from the internet. However\, from the perspective of
usability\, we all would love to freely share data and programs and use th
em. In this talk\, we shall try to understand (i) the drawbacks of current
security systems and (ii) the foundations of systems which enable us to r
ealize the goal of freely sharing data and code with untrusted partners wi
thout compromising the security.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=3569
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3569
END:VEVENT
BEGIN:VEVENT
SUMMARY:Beating Brute Force Search for QBF Satisfiability\, and Implicatio
ns for Formula Size Lower Bounds
DTSTART;VALUE=DATE-TIME:20140417T103000Z
DTEND;VALUE=DATE-TIME:20140417T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3602@cern.ch
DESCRIPTION:We give the first algorithms for the QBF Satisfiability proble
m beating brute force search. We show that the QBF satisfiability question
for CNF instances with $n$ variables\, $q$ quantifier blocks and size pol
y(n) can be solved in time 2^{n-\\Omega(n^{1/(q+1))}\, and that the QBF sa
tisfiability question for circuit instances with $n$ variables\, $q$ quant
ifier blocks and size poly(n) can be solved in time 2^{n-\\Omega(q)}. We a
lso show that improvements on these algorithms would lead to super-polynom
ial formula size lower bounds for NEXP (this is joint work with Ryan Willi
ams).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=360
2
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3602
END:VEVENT
BEGIN:VEVENT
SUMMARY:Social Algorithms and Internet Economics
DTSTART;VALUE=DATE-TIME:20140418T090000Z
DTEND;VALUE=DATE-TIME:20140418T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3615@cern.ch
DESCRIPTION:Classical algorithm design assumes that all the parameters of
the objective function is known. When that information is held privately b
y multiple agents\, we arrive in the domain of mechanism design\, where th
e parameters of a social objective function is dispersed among the strateg
ic agents who reveal this information when an algorithm can be designed to
incentivize them to do so. In the first part of the talk\, I will motivat
e the design of algorithm under incomplete information. In the latter part
\, I will discuss our recent result in the domain of quasi-linear preferen
ces with selfish valuations\, a domain that is relevant for resource shari
ng applications like mobile bandwidth or cloud computing. We show that eve
n though the valuations are less rich than that in the requirements of Rob
erts (1979)\, we are able to get the affine maximizer result with an addit
ional assumption on the social choice function and the allocation space.\n
\nThe later part of the talk is based on my recent work: https://dl.dropbo
xusercontent.com/u/26115762/site/publications.html#nath-sen14roberts\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3615
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3615
END:VEVENT
BEGIN:VEVENT
SUMMARY:Small Stretch Pairwise Spanners and D Spanners
DTSTART;VALUE=DATE-TIME:20140425T090000Z
DTEND;VALUE=DATE-TIME:20140425T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3609@cern.ch
DESCRIPTION:An additive spanner of an undirected graph is a subgraph in wh
ich the shortest distance between any pair of vertices is larger than the
corresponding distance in the original graph by at most an additive term.
This additive term is called the stretch of that spanner. Additive spanner
s are very well studied. They have applications in compact routing schemes
\, near shortest path algorithms\, approximate distance oracles\, etc. A n
atural relaxation of the above problem is where the approximation requirem
ent applies only to a subset of pairs of vertices. Subgraphs satisfying th
is relaxed requirement are called pairwise spanners. A $D$-spanner is a pa
irwise spanner for which the set of pairs of vertices to be approximated a
re all those that are at a distance at least $D$ in the original graph. It
is NP-hard to compute the sparsest additive pairwise spanner of a graph f
or any given stretch. The typical goal is to design algorithms which for s
pecific values of stretch\, construct pairwise spanners that are as sparse
as possible.\n\nIn this talk we will present a brief survey of the work d
one in the area of pairwise spanners\, the important algorithmic technique
s used in constructing them and high level ideas of some of our algorithms
for additive pairwise spanners.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=3609
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3609
END:VEVENT
BEGIN:VEVENT
SUMMARY:Conflict-free Colouring and Art Gallery Problem
DTSTART;VALUE=DATE-TIME:20140428T103000Z
DTEND;VALUE=DATE-TIME:20140428T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3598@cern.ch
DESCRIPTION:We present the weak and strong conflict-free colouring of guar
ds problem in polygon.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=3598
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3598
END:VEVENT
BEGIN:VEVENT
SUMMARY:Secure Information Flow: Formal Modelling\, Analysis and Practical
Applications
DTSTART;VALUE=DATE-TIME:20140502T090000Z
DTEND;VALUE=DATE-TIME:20140502T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3599@cern.ch
DESCRIPTION:For more than 3 decades now\, it is well understood that sever
al desirable security properties of systems can be cast in terms of inform
ation flows. In particular\, information flow control is widely regarded a
s a very useful tool for the analysis and enforcement of end-to-end securi
ty properties in the dynamic setting of mobile and distributed systems. Wi
th cloud becoming the new computing paradigm\, a lot of research effort is
directed towards overcoming the challenges in translating the simple and
elegant theory of information flow control into practical tools in the for
m of programming languages and systems. Although a variety of tools based
on information flow control already exist for addressing issues related to
specific applications\, a clean unified formal model is highly desirable.
\n\nIn this talk\, we (i) introduce the basic concepts of information flow
control\, (ii) present a process calculus enriched with information flow
notions together with algorithms for reasoning about security properties i
n this calculus and (iii) demonstrate the applications of our approach usi
ng Linux security as an example.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=3599
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3599
END:VEVENT
BEGIN:VEVENT
SUMMARY:Reliable\, Deniable\, Hidable Communication: Using Noise as Camouf
lage
DTSTART;VALUE=DATE-TIME:20140505T103000Z
DTEND;VALUE=DATE-TIME:20140505T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3610@cern.ch
DESCRIPTION:A transmitter Alice may wish to {\\it reliably} transmit a mes
sage to a receiver Bob over a binary symmetric channel (BSC)\, while simul
taneously ensuring that her transmission is {\\it deniable} from an eavesd
ropper Willie. That is\, if Willie listening to Alice's transmissions over
a {"significantly noisier"} BSC than the one to Bob\, he should be unable
to estimate even whether Alice is transmitting. Even when Alice's (potent
ial) communication scheme is publicly known to Willie (with {\\it no} comm
on randomness between Alice and Bob)\, we prove that over n channel uses A
lice can transmit a message of length O(\\sqrt{n}) bits to Bob\, deniably
from Willie. We also prove information-theoretic order-optimality of our r
esults. Finally\, we demonstrate that in the scenario when the channel noi
se parameters are unknown to all parties prior to communication\, the comm
unication scaling can be significantly improved\, to be linear in n. If ti
me permits\, will briefly describe AWGN version of the problem\, and also
a network version of the problem (the material in this talk is based on se
veral joint works\, with Pak Hou Che\, Mayank Bakshi\, Chung Chan\, Zongha
o Chen\, Alex Sprintson\, and Swanand Kadhe).\n\nThe 1 hour talk will be f
ollowed by an additional half hour with more details for those who are int
erested.\n\nBio: B.Tech. ('00)\, EE\, IIT Bombay\,\nMS/Ph.D. ('05) EE\, C
alTech\,\nPostdoctoral Associate ('06) LIDS\, MIT\,\nCurrently Associate P
rofessor\, Dept. of Information Engineering\, The Chinese University of Ho
ng Kong.\nResearch interests: Network coding and network error-correcting
algorithms\, coding theory\, steganography\, group testing\, compressive s
ensing.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3
610
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3610
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Geometric Approach to the Global Attractor Conjecture
DTSTART;VALUE=DATE-TIME:20140512T103000Z
DTEND;VALUE=DATE-TIME:20140512T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3674@cern.ch
DESCRIPTION:We tackle one of the most fascinating problems in reaction net
work theory: the Global Attractor Conjecture\, which has been an open ques
tion for four decades now. In this paper\, we do not quite prove the conje
cture\, but we prove a significant special case. We obtain a surprisingly
crisp combinatorial understanding of the dynamics. On one hand there is th
is monstrous system of Ordinary Differential Equations with arbitrary degr
ee non-linear terms on the right-hand side which seems hopeless to underst
and. On the other hand\, there is this geometric diagram\, much easier to
work with. Remarkably\, this simple geometric diagram reveals a surprising
amount of information about the non-linear Ordinary Differential Equation
system (this is joint work with Ezra Miller\, and Anne Shiu and was publi
shed in SIAM J. Appl. Dyn. Syst.\, 13(2)\, 758–797. (40 pages).\n\nhttp:
//epubs.siam.org/doi/abs/10.1137/130928170\narxiv.org/abs/1305.5303\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3674
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3674
END:VEVENT
BEGIN:VEVENT
SUMMARY:Discrete Thermodynamics
DTSTART;VALUE=DATE-TIME:20140521T103000Z
DTEND;VALUE=DATE-TIME:20140521T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3684@cern.ch
DESCRIPTION:We will explore thermodynamics in the setting of finite-space
Markov chains. Such models are interesting in three ways. From a practical
point of view\, they correspond to "coarse-graining" of real physical mod
els. From a logical point of view\, they are the finite models of the theo
ry of thermodynamics --- many proofs in this finite setting extend natural
ly to the general setting. From a pedagogical point of view\, interesting
results can be proved without requiring the prerequisites of measure theor
y and Brownian motion. In this setting\, we will explore some ideas from s
tochastic thermodynamics and information theory\, and connections between
the two.\n\nDr. Manoj Gopalkrishnan will be delivering a short course of t
wo lectures on "Discrete Thermodynamics." The lectures will be held on Wed
nesday May 21 and Friday May 23\, from 1600 to 1700 hours in D-405.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3684
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3684
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sparse Process Flexibility Designs: Is Long-Chain Really Optimal?
DTSTART;VALUE=DATE-TIME:20140522T103000Z
DTEND;VALUE=DATE-TIME:20140522T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3685@cern.ch
DESCRIPTION:Long chain refers to a directed bi-partite network with direct
ed edges from supply nodes (with fixed unit supply) to demand nodes (with
random demand) that form a hamiltonian cycle. This design was introduced i
n a seminal paper of Jordan and Graves (1995) and has been an important de
sign in process flexibility. Jordan and Graves (1995) observed empirically
that the expected performance (satisfied demand) is quite close to the co
mplete bi-partite graph for certain demand distributions. Recently\, Simch
i-Levi and Wei (2012) show that long chain maximizes the expected performa
nce among all 2-regular networks for exchangeable demand distributions.\n\
nWe study the performance of long chain in comparison to all networks with
2n edges when the assumption of 2-regularity is relaxed. We show that sur
prisingly long chain is not optimal in this class of networks even for i.i
.d. demand distributions. We present a family of instances where a disconn
ected network with 2n edges has a strictly better performance than long ch
ain.\n\nHowever\, if we restrict to connected networks with 2n arcs\, we s
how that long chain is optimal for exchangeable demand distributions. Our
proof is based on a combinatorial analysis of the structure of maximum flo
w in directed graphs and a coupling argument that reduces the comparison o
f expected performance to a sample pathwise comparison of satisfied demand
. Our analysis provides useful insights towards not just understanding the
optimality of long chain but also towards designing more general sparse f
lexibility networks (this is joint work with Antoine Desir\, Yehua Wei and
Jiawei Zhang).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=3685
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3685
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hardness of Hypergraph Coloring
DTSTART;VALUE=DATE-TIME:20140523T043000Z
DTEND;VALUE=DATE-TIME:20140523T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3694@cern.ch
DESCRIPTION:The Graph Coloring problem is to efficiently color the vertice
s of a graph on n vertices using as few colors as possible such that no ed
ge is monochromatic. All known efficient algorithms till date\, only guara
ntee a $n^\\delta$ coloring\, when the input is promised to be 3 colorable
. However what is known is that it is NP Hard to 4 color a 3 colorable gr
aph. Improving the hardness result to match the algorithms has been an ope
n question for long. However progress had been made in the hypergraph vers
ion: Given a 2 colorable hypergraph\, prove that its hard to color it usin
g Q colors. The results currently known only give hardness for upto $Q= lo
g^c n$. We give the first results that crosses the log^c barrier using the
low-degree polynomial code (aka\, the 'short code' of Barak et. al. [FOCS
2012]) and the techniques proposed by Dinur and Guruswami [FOCS 2013] to
incorporate this code for inapproximability results. In particular\, we pr
ove quasi-NP-hardness of the following problems on n-vertex hyper-graphs:
\n* Coloring a 2-colorable 8-uniform hypergraph with 2^2^Ω(√loglogn) co
lors. \n* Coloring a 4-colorable 4-uniform hypergraph with 2^2^Ω(√loglo
gn) colors. \n* Coloring a 3-colorable 3-uniform hypergraph with (logn)^Ω
(1/logloglogn) colors. \n \nReference : http://arxiv.org/abs/1311.7407\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3694
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3694
END:VEVENT
BEGIN:VEVENT
SUMMARY:Extensional Crisis and Proving Identity
DTSTART;VALUE=DATE-TIME:20140526T103000Z
DTEND;VALUE=DATE-TIME:20140526T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3696@cern.ch
DESCRIPTION:Extensionality axioms are common when reasoning about data col
lections\, such as arrays and functions in program analysis\, or sets in m
athematics. An extensionality axiom asserts that two collections are equal
if they consist of the same elements at the same indices.\n\nUsing extens
ionality is often required to show that two collections are equal. A typic
al example is the set theory theorem \\forall x)(\\forall y) x \\union y =
y \\union x. Interestingly\, while humans have no problem with proving su
ch set identities using extensionality\, they are very hard for superposit
ion theorem provers because of the calculi they use.\n\nIn this work we sh
ow how addition of a new inference rule\, called extensionality resolution
\, allows first-order theorem provers to easily solve problems no modern f
irst-order theorem prover can solve. We illustrate this by running the VAM
PIRE theorem prover with extensionality resolution on a number of set theo
ry and array problems. Extensionality resolution helps VAMPIRE to solve pr
oblems from the TPTP library of first-order problems that were never solve
d before by any prover.\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=3696
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3696
END:VEVENT
BEGIN:VEVENT
SUMMARY:Multidimensional Scaling and Isomap
DTSTART;VALUE=DATE-TIME:20140530T090000Z
DTEND;VALUE=DATE-TIME:20140530T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3700@cern.ch
DESCRIPTION:Multidimensional scaling (MDS) is used to visualize pairwise s
imilarity between objects. Given a distance matrix between objects\, we wi
ll see how MDS can be used to generate points in some vector space which r
espect these distances. We will also talk about how this can be extended t
o a non-linear dimensionality reduction technique known as the Isomap.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3700
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3700
END:VEVENT
BEGIN:VEVENT
SUMMARY:Automated Test Generation Using Concolic Testing
DTSTART;VALUE=DATE-TIME:20140603T060000Z
DTEND;VALUE=DATE-TIME:20140603T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3690@cern.ch
DESCRIPTION:In this talk\, I will describe concolic testing\, also known a
s directed automated random testing (DART) or dynamic symbolic execution\,
an efficient way to automatically and systematically generate test inputs
for programs. Concolic testing uses a combination of runtime symbolic exe
cution and automated theorem proving techniques to automatically generate
non-redundant and exhaustive test inputs. Specifically\, concolic testing
performs symbolic execution along a concrete execution path\, generates a
logical formula denoting a constraint on the input values\, and solves a c
onstraint to generate new test inputs that would execute the program along
previously unexplored paths. Concolic testing has inspired the developmen
t of several industrial and academic automated testing and security tools
such as PEX\, SAGE\, and YOGI at Microsoft\, Apollo at IBM\, Conbol at Sam
sung\, and CUTE\, jCUTE\, CATG\, Jalangi\, SPLAT\, BitBlaze\, jFuzz\, Oasi
s\, and SmartFuzz in academia. A central reason behind the wide adoption o
f concolic testing is that\, while concolic testing uses program analysis
and automated theorem proving techniques internally\, it exposes a testing
usage model that is familiar to most software developers.\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=3690
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3690
END:VEVENT
BEGIN:VEVENT
SUMMARY:Testing\, Debugging\, and Precision-tuning of Large-scale Parallel
and Floating-point Programs
DTSTART;VALUE=DATE-TIME:20140604T103000Z
DTEND;VALUE=DATE-TIME:20140604T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3691@cern.ch
DESCRIPTION:Testing and debugging scientific applications is a difficult t
ask\, in particular for large-scale parallel programs and floating-point p
rograms. Parallel programs often exhibit incorrect behaviors due to uninte
nded interference among multiple threads. Such concurrency bugs are often
difficult to pinpoint because they typically happen under very specific th
read interleavings. Floating-point programs are difficult to write\, test\
, and debug because such programs can produce incorrect results due to a v
ariety of numerical errors that can happen during execution. In this talk\
, Professor Koushik will describe a couple of program analysis and debuggi
ng techniques to access the correctness of parallel programs written using
hybrid parallelism and floating-point datatypes.\n\nAbout the Speaker: Pr
ofesor Koushik Sen is an Associate Professor in the Department of Electric
al Engineering and Computer Sciences at the University of California\, Ber
keley. His research interest lies in Software Engienering\, Programming La
nguages\, and Formal Methods\, and is best known for his work on "DART: Di
ected Automated Random Testing" and concolic testing. He has been bestowed
with many awards including the NSF CAREER Award in 2008\, Haifa Verificat
ion Conference (HVC) Award in 2009\, IFIP TC2 Manfred Paul Award for Excel
lence in Software: Theory and Practice in 2010\, and Sloan Foundation Fell
owship in 2011\, to mention a few.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=3691
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3691
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Analysis and Design of Network Congestion Games
DTSTART;VALUE=DATE-TIME:20140605T063000Z
DTEND;VALUE=DATE-TIME:20140605T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3698@cern.ch
DESCRIPTION:In many applications\, users strategically choose paths in a n
etwork to minimize the congestion they face. Examples of such applications
are road traffic\, data networks\, and machine scheduling. Network conges
tion games model this strategic behaviour of the users\, and are used to u
nderstand and predict the impact of this behaviour on congestion in the ne
twork.\n\nIn this talk\, I first introduce network congestion games and pr
esent results for fundamental properties of these games. I then consider a
natural design problem: to increment capacity in the network under a fixe
d budget\, so that the average congestion for the strategic users is minim
ized. This problem is widely studied in transportation research. Despite t
his\, there are very few guarantees for polynomial-time algorithms. I will
present both algorithms and hardness results for this network improvement
problem in different network topologies\, and describe a number of relate
d problems as directions for future research.\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=3698
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3698
END:VEVENT
BEGIN:VEVENT
SUMMARY:Low-Complexity Scheduling Policies for Achieving Throughput and De
lay Optimality in Multi-channel (OFDM) Downlink Systems
DTSTART;VALUE=DATE-TIME:20140611T063000Z
DTEND;VALUE=DATE-TIME:20140611T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3729@cern.ch
DESCRIPTION:The dramatic increases in demands from multimedia applications
have put an enormous strain on the current cellular system infrastructure
. This has resulted in significant research and development efforts on 4G
multi-channel wireless cellular systems (e.g.\, LTE and WiMax) that target
new ways to achieve higher data rates\, lower latencies\, and a much bett
er user experience. Further\, in these multi-channel systems\, such as OFD
M\, the Transmission Time Interval (TTI)\, within which the scheduling dec
isions need to be made\, is typically on the order of a few milliseconds.
On the other hand\, there are hundreds of orthogonal channels that can be
allocated to different users. Hence\, many decisions have to be made withi
n a short scheduling cycle\, which means that it is critical that scheduli
ng policies must have low complexity. Thus\, a major challenge in the deve
lopment of next generation wireless networks is to design scheduling polic
ies that can simultaneously provide high throughput\, low delay\, and low
complexity. In this talk\, I will present a unifying framework for designi
ng low-complexity scheduling policies in the downlink of multi-channel (e.
g.\, OFDM-based) wireless networks that can provide optimal performance in
terms of both throughput and delay. We first develop new easy-to verify s
ufficient conditions for rate-function delay-optimality in the many-channe
l many-user asymptotic regime\, and for throughput optimality in general (
non-asymptotic) settings. The sufficient conditions enable us to prove rat
e-function delay-optimality for a class of Oldest Packets First (OPF) poli
cies and throughput optimality for a large class of Maximum Weight in the
Fluid limit (MWF) policies. While a recently developed scheduling policy i
s both throughput-optimal and rate-function delay-optimal\, it has a very
high complexity of $O(n^5)$\, where n is the number of channels or users\,
rendering it impractical. By intelligently combining policies from the cl
asses of OPF policies and MWF policies\, we design hybrid policies that ha
ve a low complexity of $O(n^{2.5} log n)$\, and are yet both throughput an
d rate-function delay optimal. We further develop two simpler greedy polic
ies that are throughput-optimal and have are near delay-optimal. The effic
acy of our schemes and comparisons with the state-of-the-art are also illu
strated through simulations.\n\nBio: Ness Shroff received his Ph.D. degree
in Electrical Engineering from Columbia University in 1994. He joined Pur
due university immediately thereafter as an Assistant Professor. At Purdue
\, he became Full Professor of ECE in 2003 and director of CWSA in 2004\,
a university-wide center on wireless systems and applications. In July 200
7\, he joined The Ohio State University\, where he holds the Ohio Eminent
Scholar endowed chair in Networking and Communications\, in the department
s of ECE and CSE. From 2009-2012\, he served as a Guest Chaired professor
of Wireless Communications at Tsinghua University\,Beijing\, China\, and c
urrently holds an honorary Guest professor at Shanghai Jiaotong University
in China. His research interests span the areas of communication\, social
\, and cyberphysical networks. He is especially interested in fundamental
problems in the design\, control\, performance\, pricing\, and security of
these networks. He currently serves as editor-at-large of IEEE/ACM Trans.
on Networking\, and on the editorial board of IEEE Trans. on Networked Co
ntrol systems\, IEEE Network Magazine\, and the Networking Science journal
. Dr. Shroff is a Fellow of the IEEE and an NSF CAREER awardee. His work h
as received numerous best paper awards for his research\, e.g.\, at IEEE I
NFOCOM 2008\, IEEE INFOCOM 2006\, Journal of Communication and Networking
2005\, and Computer Networks 2003 (his papers also received runner-up awar
ds at IEEE INFOCOM 2005 and IEEE INFOCOM 2013)\, and also student best pap
er awards (from all papers whose first author is a student) at IEEE WiOPT
2013\, IEEE WiOPT 2012\, and IEEE IWQoS 2006. In 2014\, he received the IE
EE INFOCOM Achievement award for seminal contributions to scheduling and r
esource allocation in wireless networks.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=3729
LOCATION:Colaba Campus AG-66 (Leecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3729
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computer and Its Applications
DTSTART;VALUE=DATE-TIME:20140612T103000Z
DTEND;VALUE=DATE-TIME:20140612T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3733@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3733
LOCATION:Colaba Campus AG-66 (Leecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3733
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computer Security
DTSTART;VALUE=DATE-TIME:20140613T090000Z
DTEND;VALUE=DATE-TIME:20140613T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3721@cern.ch
DESCRIPTION:Protecting data in computer systems from being misused is a ba
sic requirement. In this talk\, i shall (i) introduce the problem of prote
ction in operating systems\, (ii) discuss currently employed techniques an
d their shortcomings\, (iii) present some advanced techniques that better
address the problem\, and (iv) discuss the challenges in successfully empl
oying these techniques for practical systems.\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=3721
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3721
END:VEVENT
BEGIN:VEVENT
SUMMARY:Minimizing the Number of Unhappy Singles: An Improved Approximatio
n Algorithm for the Stable Marriage Problem With One-sided Ties
DTSTART;VALUE=DATE-TIME:20140613T060000Z
DTEND;VALUE=DATE-TIME:20140613T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3728@cern.ch
DESCRIPTION:We consider the problem of computing a large stable matching i
n a bipartite graph G = (A U B\, E) where each vertex ranks its neighbors
in an order of preference\, perhaps involving ties.The goal is to compute
a large stable matching. It is known that computing a maximum size stable
matching is APX-hard. We present an improved approximation algorithm for t
he case when ties are present only in the preference lists of vertices in
B. This case is also APX-hard and the current best approximation ratio kno
wn here is 25/17\, we show a simple linear time algorithm that achieves an
approximation ratio of 22/15.\n\nhttps://indico.tifr.res.in/indico/confer
enceDisplay.py?confId=3728
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3728
END:VEVENT
BEGIN:VEVENT
SUMMARY:Implementation of Topic models with Variational Inference
DTSTART;VALUE=DATE-TIME:20140616T103000Z
DTEND;VALUE=DATE-TIME:20140616T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3734@cern.ch
DESCRIPTION:In today's age we have an abundance of massive data available
online. This creates fantastic opportunities for Statistical learning rese
archers to analyze data and estimate ongoing patterns. Analyzing this copi
ous amount of data requires us to make certain assumptions about the model
. We present a technique called Topic models. Topic models are probabilist
ic models of text. Here we assume that data exhibits a recurring pattern o
f sets of semantically related words. These sets of semantically related w
ords are called topics. With the help of Topic models we extract and uncov
er these patterns. Topic models essentially encode our assumptions into la
tent variables\, and then infer a probability distribution over these late
nt variables. But often inferring this joint distribution is intractable.
We will present techniques which approximate the intractability of the pro
blem. Variational inference is such an approximation algorithm. We illustr
ate the most commonly used Topic model called Latent Dirichlet Allocation
and use Variational Inference techniques to approximate the posterior dist
ribution.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId
=3734
LOCATION:Colaba Campus AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3734
END:VEVENT
BEGIN:VEVENT
SUMMARY:Small Diameter Decomposition
DTSTART;VALUE=DATE-TIME:20140620T104500Z
DTEND;VALUE=DATE-TIME:20140620T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3741@cern.ch
DESCRIPTION:In this talk we study the problem of partitioning a plane set
into minimum number of parts with smaller diameters.\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=3741
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3741
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Computational Complexity of Data Flow Analysis Over Finite
Bounded Meet-semilattices
DTSTART;VALUE=DATE-TIME:20140624T100000Z
DTEND;VALUE=DATE-TIME:20140624T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3743@cern.ch
DESCRIPTION:In order to perform global code optimizations\, the compiler n
eeds to gather information about a program. It may want to know whether a
value computed will be used later during the execution of the program (Liv
e Variables Analysis) or whether during the execution\, the result of an e
xpression computation can be reused at several places (Available Expressio
ns Analysis). Data flow analysis is an abstraction of these and several ot
her similar analyses. This information can be gathered by computing the Me
et Over all Paths (MOP) solution for Data Flow Analysis. However\, if the
underlying semilattice is infinite\, computing the MOP solution is undecid
able. So as a safe approximation\, we settle with the Maximum Fixed Point
(MFP) solution which is polynomial time computable for semilattices of fin
ite height.\n\nIn this talk\, we consider Data Flow Analysis over monotone
data flow frameworks with a finite bounded meet-semilattice. We show that
the problem of computing the MFP solution is P-complete. This shows that
the problem can not have efficient parallel algorithms unless P = NC. We w
ill also show that the problem of computing the MOP solution is NL-complet
e (and hence efficiently parallelizable) when the semilattice is finite. T
hese results appear in contrast with the fact that computing the MOP solut
ion is significantly harder than the MFP solution in general.\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=3743
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3743
END:VEVENT
BEGIN:VEVENT
SUMMARY:Private Information Retrieval (PIR)
DTSTART;VALUE=DATE-TIME:20140627T090000Z
DTEND;VALUE=DATE-TIME:20140627T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3748@cern.ch
DESCRIPTION:Informally\, the problem of PIR studies how to access a record
from a server without the server knowing what record is being retrived. F
ormally\, the problem is defined as follows: the server has a binary strin
g (X^n) of length n and the user wants to retrieve the Yth bit X_Y in such
a way that the server does not learn Y at all. In this talk\, we will pro
ve optimality (communication complexity) of a trivial scheme where the ser
ver sends the whole string to the user. A better scheme is possible if we
replicate the same string in two servers and we will see such a scheme. We
also discuss a variant of PIR\, called symmetric PIR\, where we put one m
ore constraint that the user also shouldn't learn anything about the datab
ase other than the output. We will prove that for any number k>=1 of serve
rs\, if no two servers communicate\, then symmetric PIR can't be done.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3748
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3748
END:VEVENT
BEGIN:VEVENT
SUMMARY:Communication Complexity
DTSTART;VALUE=DATE-TIME:20140704T090000Z
DTEND;VALUE=DATE-TIME:20140704T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3752@cern.ch
DESCRIPTION:Alice and Bob want to compute jointly/collaboratively a functi
on. They are both super-computers. However\, part of the input is with Ali
ce and the other part is with Bob. How many bits do they *need* to communi
cate with each other\, in the worst case\, so that they can compute the fu
nction?\n\nIt turns out that the above puzzle and some of its variants hav
e deep connection with all sorts of important questions: can we process la
rge data using small memory? can we have efficient data-structures for a v
ariety of problems? does there exist parallel algorithms for factoring?...
and many others...\n\nWe will introduce the basic model of communication
and briefly go through some of these exciting connections.\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=3752
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3752
END:VEVENT
BEGIN:VEVENT
SUMMARY:Topology of a Randomly Evolving Erdos Renyi Graph
DTSTART;VALUE=DATE-TIME:20140711T090000Z
DTEND;VALUE=DATE-TIME:20140711T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3759@cern.ch
DESCRIPTION:Mathew Kahle and Elizabeth Meckes recently established interes
ting results concerning the topology of the clique complex $X(n\,p)$ on an
Erdos Renyi graph $G(n\,p).$ Specifically\, they showed that\, if $p = n^
{\\alpha}$ with $\\alpha \\in (-1/k\, -1/(k + 1))$ for some positive integ
er $k\,$ then asymptotically\, i.e.\, as $n \\rightarrow \\infty\,$ every
Betti number $\\beta_j$ of $X(n\,p)\,$ except for the $k-$th one\, vanishe
s. Further\, for the choice of $p$ as above\, $\\beta_k$ of $X(n\,p)$ foll
ows a central limit theorem\, i.e.\, $(\\beta_k - \\mathbb{E}[\\beta_k])/\
\sqrt{Var(\\beta_k)}$ is asymptotically Gaussian.\n\nIn this talk\, we wil
l consider a randomly evolving Erdos Renyi graph $G(n\, p\, t)$ and study
how its topology varies with time $t.$ Specifically\, we will prove that i
f p is chosen as above\, then the process $(\\beta_k(t) - \\mathbb{E}[\\be
ta_k(t)])/\\sqrt{Var[\\beta_k(t)]}$ is asymptotically an Ornstein-Uhlenbec
k process. That is\, the k-th Betti number asymptotically behaves like a s
tationary Gaussian Markov process with an exponentially decaying covarianc
e function.\n\nI will NOT assume any prerequisites for this talk.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3759
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3759
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Structure of Boolean Functions With Small Spectral Norm
DTSTART;VALUE=DATE-TIME:20140718T090000Z
DTEND;VALUE=DATE-TIME:20140718T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3786@cern.ch
DESCRIPTION:Let f: F_2^n -> {+1\, -1} be a Boolean function with the first
Fourier norm A and Fourier sparsity s. We will prove that there is an aff
ine subspace of the vector space F_2^n\, of dimension O(A)\, on which the
f is constant. If time permits\, we will prove that f has a parity decisio
n tree of depth O(\\sqrt{s}). These results are by Tsang et al which impr
oves on earlier results by Shpilka et al.\n\nReferences: \nAmir Shpilka\,
Avishay Tal\, Ben lee Volk. On the Structure of Boolean Functions with Sma
ll Spectral Norm. ITCS 2014\nHing Yin Tsang\, Chung Hoi Wong\, Ning Xie\,
Shengyu Zhang. Fourier sparsity\, spectral norm and the log rank conjectur
e. FOCS 2013\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?con
fId=3786
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3786
END:VEVENT
BEGIN:VEVENT
SUMMARY:Test-Driven Detection and Repair of Data Races in Structured Paral
lel Programs
DTSTART;VALUE=DATE-TIME:20140730T060000Z
DTEND;VALUE=DATE-TIME:20140730T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3783@cern.ch
DESCRIPTION:In this talk\, we focus on a programmability challenge for par
allel computing that can be captured by the following question: how easy o
r difficult is it to identify and repair bugs when developing a parallel p
rogram? Our claim is that structured parallel programming greatly simplifi
es the task of writing correct and efficient parallel programs. We support
this claim by first introducing a family of structured parallel programmi
ng models developed in the Habanero Extreme Scale Software Research projec
t [1] at Rice University\, and discussing semantic guarantees (e.g.\, dead
lock freedom\, datarace freedom\, determinism\, serial elision) for differ
ent subsets of these structured parallel programs. Then we discuss recent
results for test-driven datarace detection [2] and repair [3] using struct
ured parallelism. The results in [2\,3] were demonstrated using the Habane
ro-Java language on single multicore nodes. If time permits\, we will brie
fly discuss results for cluster-level parallelism using a novel integratio
n of task parallelism with MPI in Habanero-C [4]\, as well as new research
directions being pursued in the Habanero project (this work was done in c
ollaboration with members of the Habanero Extreme Scale Software Research
group).\n\nBio: Vivek Sarkar is Professor and Chair of Computer Science at
Rice University. He conducts research in multiple aspects of parallel sof
tware including programming languages\, program analysis\, compiler optimi
zations and runtimes for parallel and high performance computer systems. H
e currently leads the Habanero Extreme Scale Software Research project at
Rice University\, and serves as Associate Director of the NSF Expeditions
Center for Domain-Specific Computing. Prior to joining Rice in 2007\, Vive
k was Senior Manager of Programming Technologies at IBM Research. His prio
r research projects include the X10 programming language\, the Jikes Resea
rch Virtual Machine for the Java language\, the ASTI optimizer used in IBM
’s XL Fortran product compilers\, the PTRAN automatic parallelization sy
stem\, and profile-directed partitioning and scheduling of Sisal programs.
In 1997\, he was on sabbatical as a visiting associate professor at MIT\,
where he was a founding member of the MIT Raw multicore project. Vivek be
came a member of the IBM Academy of Technology in 1995\, the E.D. Butcher
Chair in Engineering at Rice University in 2007\, and was inducted as an A
CM Fellow in 2008. He holds a B.Tech. degree from the Indian Institute of
Technology\, Kanpur\, an M.S. degree from University of Wisconsin-Madison\
, and a Ph.D. from Stanford University. Vivek has been serving as a member
of the US Department of Energy’s Advanced Scientific Computing Advisory
Committee (ASCAC) since 2009.\n\nReferences: [1] Habanero Extreme Scale S
oftware Research project (http://habanero.rice.edu) [2] Scalable and Preci
se Dynamic Data Race Detection for Structured Parallelism. Raghavan Raman\
, Jisheng Zhao\, Vivek Sarkar\, Martin Vechev\, Eran Yahav. PLDI 2012. [3]
Test-Driven Repair of Data Races in Structured Parallel Programs. Rishi S
urendran\, Raghavan Raman\, Swarat Chaudhuri\, John Mellor-Crummey\, Vivek
Sarkar. PLDI 2014. [4] Integrating Asynchronous Task Parallelism with MPI
. Sanjay Chatterjee\, Sagnak Tasirlar\, Zoran Budimlic\, Vincent Cave \, M
ilind Chabbi\, Max Grossman\, Yonghong Yan\, Vivek Sarkar. IPDPS 2013.\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3783
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3783
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Size of Kakeya Sets in Finite Fields
DTSTART;VALUE=DATE-TIME:20140801T090000Z
DTEND;VALUE=DATE-TIME:20140801T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3812@cern.ch
DESCRIPTION:A Kakeya set is a subset of [image: F^n]\, where [image: F] is
a finite field of [image: q] elements\, that contains a line in every dir
ection. What can we say about the size of this set? How large the size of
the set must be?\n\nI will be discussing the paper: On the size of Kakeya
sets in finite fields by Zeev Dvir. In the paper\, a beautiful application
of 'polynomial method' is used to give a lower bound on the size of the K
akeya Set.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=3812
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3812
END:VEVENT
BEGIN:VEVENT
SUMMARY:Testing the Manifold Hypothesis
DTSTART;VALUE=DATE-TIME:20140804T090000Z
DTEND;VALUE=DATE-TIME:20140804T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3795@cern.ch
DESCRIPTION:We are confronted with very high dimensional data sets. As a r
esult\, methods of dealing with high dimensional data have become prominen
t. One geometrically motivated approach for analyzing data is called manif
old learning. The underlying hypothesis of this subfield of machine learni
ng is that high dimensional data tend to lie near a low dimensional manifo
ld. However\, the basic question of understanding when data lies near a ma
nifold is poorly understood. I will describe joint work with Charles Feffe
rman and Sanjoy Mitter on developing a provably correct algorithm to test
this hypothesis using i.i.d samples from an arbitrary distribution support
ed in the unit ball in a Hilbert space.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=3795
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3795
END:VEVENT
BEGIN:VEVENT
SUMMARY:Randomized Interior Point Methods for Sampling and Optimization
DTSTART;VALUE=DATE-TIME:20140805T090000Z
DTEND;VALUE=DATE-TIME:20140805T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3796@cern.ch
DESCRIPTION:Interior point methods are algorithms that optimize convex fun
ctions over high dimensional convex sets. From one point of view\, an inte
rior point method first equips a convex set with a Riemannian metric and t
hen performs a steepest descent to minimize the objective on the resulting
Riemannian manifold. We will describe a randomized variant of an interior
point method known as ``the affine scaling algorithm" introduced by I.I.D
ikin. This variant corresponds to a natural random walk on the same manifo
ld on which affine scaling would perform steepest descent. We discuss appl
ications to sampling and optimization and prove polynomial bounds on the m
ixing time of the associated Markov Chain. This talk includes work done i
n collaboration with Ravi Kannan and Alexander Rakhlin.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=3796
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3796
END:VEVENT
BEGIN:VEVENT
SUMMARY:Convex Relative Entropy Decay in Markov Chains
DTSTART;VALUE=DATE-TIME:20140805T103000Z
DTEND;VALUE=DATE-TIME:20140805T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3813@cern.ch
DESCRIPTION:Consider an irreducible continuous time Markov chain with a fi
nite or a countably infinite number of states and admitting a unique stati
onary probability distribution. The relative entropy of the distribution o
f the chain at any time with respect to the stationary distribution is a m
onotonically decreasing function of time. It is interesting to ask if this
function is convex. We discuss this question for finite Markov chains and
for Jackson networks\, which are a class of countable state Markov chains
of interest in modelingnetworks of queues (joint work with Varun Jog).\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3813
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3813
END:VEVENT
BEGIN:VEVENT
SUMMARY:Randomized Load Balancing in Large Bandwidth Sharing Systems
DTSTART;VALUE=DATE-TIME:20140807T103000Z
DTEND;VALUE=DATE-TIME:20140807T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3790@cern.ch
DESCRIPTION:Processor sharing models occur in a wide variety of situations
for example in models of internet bottlenecks. They are good models for b
andwidth sharing as well as being solutions to NUM for logarithmic utiliti
es. In addition they possess the desirable stochastic property of the stat
ionary distribution being insensitive to the service time distribution. In
this talk I will discuss new advances in understanding and characterizing
the behavior of randomized routing to PS servers that are heterogeneous i
n terms of their server speeds.\n\nWe will rst discuss the so-called Power
-of-two rule in the homogeneous case of identical servers where routing to
the least occupied server amongst two randomly chosen servers results in
a very low server occupancy and a so-called propagation of chaos or asympt
otic independence. There were no known characterizations about the heterog
eneous case. In the heterogeneous case we will see that the stability regi
on for randomized routing is strictly included in the maximal stability re
gion that can be achieved by state independent routing. Therefore the aver
age sojourn of tasks can be longer in randomized routing in heterogeneous
systems. When the system is stable we completely characterize the steady-s
tate behavior of the server occupancies and show that it exhibits super-ex
ponential decay and asymptotic independence among servers. To overcome the
reduction in the stability region we show that a combination of state ind
ependent routing (biased sampling) to a server class combined with JSQ wit
hin the class recovers the stability region as well as the benets of small
server occupancies.\n\nThe techniques are based on a mean eld analysis an
d an ansatz based on propagation of chaos that then establishes the asympt
otic independence between servers (joint work with Arpan Mukhopadhyay (Wat
erloo).\n\nBiography: The speaker was educated at the Indian Institute of
Technology\, Bombay (B.Tech\, 1977)\, Imperial College\, London (MSc\, DIC
\, 1978) and obtained his PhD under A. V. Balakrishnan at UCLA in 1983. He
is currently a University Research Chair Professor in the Dept. of ECE at
the University of Waterloo\, Ont.\, Canada where he has been since Septem
ber 2004. Prior to this he was Professor of ECE at Purdue University\, Wes
t Lafayette\, USA. Since 2012 he is a D.J. Gandhi Distinguished Visiting P
rofessor at the Indian Institute of Technology\, Bombay\, India. He is a F
ellow of the IEEE and the Royal Statistical Society. He is a recipient of
the INFOCOM 2006 Best Paper Award and was runner-up for the Best Paper Awa
rd at INFOCOM 1998. His research interests are in modeling\, control\, and
performance analysis of both wireline and wireless networks\, and in appl
ied probability and stochastic analysis with applications to queueing\, lt
ering\, and optimization.\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=3790
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3790
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Basic Linear Algebraic Techniques for Proving Lower Bounds
DTSTART;VALUE=DATE-TIME:20140808T090000Z
DTEND;VALUE=DATE-TIME:20140808T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3833@cern.ch
DESCRIPTION:In this talk I will discuss some linear algebraic methods to p
rove lower bounds. In particular I will discuss Graham Pollak theorem\, a
graph theoretic result which gives us a lower bound on the number of bipar
tite cliques needed to cover a complete graph. Also\, I will talk about Ra
zbarov's result that threshold functions cannot be approximated by small d
egree polynomials and may be more if time permits.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=3833
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3833
END:VEVENT
BEGIN:VEVENT
SUMMARY:Higher-order Fourier Analysis and Applications
DTSTART;VALUE=DATE-TIME:20140811T043000Z
DTEND;VALUE=DATE-TIME:20140811T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3837@cern.ch
DESCRIPTION:Regularity is a notion of “pseudorandomness” that allows o
ne to decompose a given object into a collection of simpler objects which
appear random according to certain statistics. The famous regularity lemma
of Szemeredi [Sze75\, Sze78] says that any dense graph can be partitioned
into a collection of bounded number of “pseudorandom” bipartite graph
s. The Szemeredi regularity lemma has numerous applications in combinatori
cs and property testing.\n\nIn a sequence of developments stemming from Go
wers' proof of Szemeredi's theorem\, Green and Tao introduced a notion of
regularity for a collection of polynomials. Variants of these ideas were f
amously used by Green and Tao to prove that the primes contain arbitrarily
long arithmetic progressions. Over finite fields\, the theory extends pre
viously used concepts in theoretical computer science\, such as low-biased
random variables and Fourier analysis over finite fields.\n\nIn this talk
\, we will survey recent developments in the area\, loosely termed "higher
-order Fourier analysis". In joint work with Fischer\, H. Hatami\, P. Hata
mi\, and Lovett\, we developed certain bounds for exponential sums of poly
nomials [BFL13\, BFH+13] that are useful in the area of property testing.
These results used a non-constructive form of the regularity lemma. Subseq
uently\, in joint work with P. Hatami and Tulsiani [BHT13]\, we devised an
algorithmic regularity lemma for polynomials. This can be used [B14] to g
ive new polynomial time algorithms for problems such as factoring low-degr
ee multivariate polynomials over prime order fields and deciding\, for fix
ed $p\, d$ and $r$\, whether a given d-dimensional tensor over $F_p$ has r
ank at most $r$.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=3837
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3837
END:VEVENT
BEGIN:VEVENT
SUMMARY:New Compact Linear Programming Formulations for Choice Network Rev
enue Management
DTSTART;VALUE=DATE-TIME:20140812T090000Z
DTEND;VALUE=DATE-TIME:20140812T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3794@cern.ch
DESCRIPTION:We consider the network RM problem with customer choice\, whic
h incorporates customer purchase behavior as a function of the offered pro
ducts. The optimization problem is a stochastic dynamic program and is int
ractable. A certainty-equivalence relaxation of the dynamic program\, call
ed the choice deterministic linear program (CDLP) is usually used to gener
ate dynamic controls. It is known CDLP has a compact formulation for the m
ultinomial-logit (MNL) model of customer choice. Our objective is to obtai
n a tighter bound than CDLP while retaining the appealing properties of a
compact linear programming representation. To this end\, it is natural to
consider the affine relaxation of the dynamic program. We first show that
the affine relaxation is NP-complete even for a single-segment MNL model.
Nevertheless\, by analyzing the affine relaxation we derive new linear pro
grams that approximate the dynamic programming value function better than
CDLP\, provably between the CDLP value and the affine relaxation. We give
a bound on the gap between these tractable relaxations and the (NP-hard) a
ffine relaxation. This bound pinpoints the range of capacities where impro
vements are possible. We give extensions to the case with multiple custome
r segments and nested-logit model of choice. Finally we perform extensive
numerical comparisons on the various bounds to evaluate their peformance (
joint work with K. Talluri).\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=3794
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3794
END:VEVENT
BEGIN:VEVENT
SUMMARY:Security Protocols: An Information Flow Perspective
DTSTART;VALUE=DATE-TIME:20140813T103000Z
DTEND;VALUE=DATE-TIME:20140813T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3839@cern.ch
DESCRIPTION:There has been a tremendous increase in the development and us
e of networked and distributed systems in the past three decades. Security
protocols are small programs that aim at securing communications over a p
ublic network like the Internet. E-commerce is now an integral part of our
daily lives\, and security protocols play a major role in securing these
transactions online. Protocols are commonly described using an informal no
tation that only gives an intended trace involving the honest principals.
These protocol descriptions/narrations often leave many properties of a pr
otocol unspecified. In particular\, there is no way to determine the initi
al conditions or assumptions about shared information\, nor can we see und
er what conditions the principals should respond to messages.\n\nDesign of
protocols is in itself a challenging task\, but is not the central theme
of this talk. Rather\, we argue that viewing a protocol as an exchange of
information among the intended stakeholders of the system\, and applying t
he concepts of information flow security for enriching the protocol specif
ications helps the designer to make his intentions explicit. We illustrate
our approach using the Needham-Schroeder protocol as an example.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3839
LOCATION:Colaba Campus AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3839
END:VEVENT
BEGIN:VEVENT
SUMMARY:Regenerative sequence Monte Carlo
DTSTART;VALUE=DATE-TIME:20140814T103000Z
DTEND;VALUE=DATE-TIME:20140814T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3811@cern.ch
DESCRIPTION:In this talk we introduce a new Monte Carlo (mc) procedure to
consider estimating integrals w.r.t. infinite measures. The iid mc or mcmc
are not applicable here. We consider both point estimation and interval e
stimation (this is joint work with Vivek Roy of Iowa State university).\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3811
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3811
END:VEVENT
BEGIN:VEVENT
SUMMARY:Improved Counting Relative to Pseudorandom Graphs
DTSTART;VALUE=DATE-TIME:20140814T043000Z
DTEND;VALUE=DATE-TIME:20140814T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3842@cern.ch
DESCRIPTION:A graph is `pseudorandom' if it appears random according to ce
rtain statistics. Recently\, Conlon\, Fox and Zhao proved a counting lemma
\, counting small graphs in $\\varepsilon$-regular subgraphs of sparse pse
udorandom graphs. This counting lemma has many important applications such
as sparse pseudorandom analogues of Tur\\’an’s Theorem\, Ramsey’s T
heorem and the graph removal lemma.\n\nOne key ingredient for the proof of
their counting lemma is a regularity inheritance lemma\, which states tha
t for most vertices in an $\\varepsilon$-regular subgraph of a pseudorando
m graph\, the neighbourhoods of this vertex form an $\\varepsilon’$-regu
lar graph. We improve this regularity inheritance lemma\, so that it now a
pplies to graphs with weaker pseudorandomness conditions. This implies an
improved counting lemma relative to these pseudorandom graphs (based on jo
int work with Peter Allen\, Julia Boettcher\, Maya Stein).\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=3842
LOCATION:Colaba Campus AG-66 (Leecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3842
END:VEVENT
BEGIN:VEVENT
SUMMARY:Small Value Parallel Repetition for General Games
DTSTART;VALUE=DATE-TIME:20140821T103000Z
DTEND;VALUE=DATE-TIME:20140821T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3856@cern.ch
DESCRIPTION:We prove a parallel repetition theorem for general games with
value tending to 0. Previously Dinur and Steurer proved such a theorem for
the special case of projection games. We use information theoretic techni
ques in our proof. Our proofs also extend to the high value regime (value
close to 1) and provide alternate proofs for the parallel repetition theor
ems of Holenstein and Rao for general and projection games respectively. W
e also extend the example of Feige and Verbitsky to show that the small-va
lue parallel repetition bound we obtain is tight. Our techniques are eleme
ntary in that we only need to employ basic information theory and discrete
probability in the small-value parallel repetition proof (this is joint w
ork with Mark Braverman).\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=3856
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3856
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Kemnitz Conjecture and its Generalization for Z_p^3 Where p is
a Prime Number
DTSTART;VALUE=DATE-TIME:20140822T083000Z
DTEND;VALUE=DATE-TIME:20140822T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3867@cern.ch
DESCRIPTION:I will sketch the proof as given by C. Reiher.\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=3867
LOCATION:Colaba Campus D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3867
END:VEVENT
BEGIN:VEVENT
SUMMARY:Monte Carlo Based Methods for Pricing American Options
DTSTART;VALUE=DATE-TIME:20140826T090000Z
DTEND;VALUE=DATE-TIME:20140826T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3868@cern.ch
DESCRIPTION:In this thesis\, we develop computational algorithms in a numb
er of different regimes for pricing financial American options. As is well
known\, American option is a contract between two parties that gives the
buyer of the option the right (but not the obligation) to buy or sell an u
nderlying asset at a certain strike price any time before expiration date
(also called maturity) of the option. Their pricing problem reduces to sol
ving the optimal stopping problem for a Markov process. Monte Carlo is a p
referred tool for solving this problem when the underlying state space is
high dimensional.\n\nWe study three aspects of the pricing problem in this
thesis. First\, we propose a new simulation method to price American opti
on written on a single underlying asset under the popular stochastic volat
ility model. We derive the optimal exercise boundary approximation and use
it in conjunction with Monte Carlo to find estimator for true American op
tion price. We also derive option price approximation to use it to form ma
rtingale control variates for proposed estimator and demonstrate considera
ble variance reduction.\n\nIn addition\, for high-dimensional American opt
ions\, under general model settings\, we propose a new simulation method b
ased on the nearest neighbor estimation technique. We derive the optimal m
ean square error (MSE) convergence rate for the proposed estimator and stu
dy the impact of underlying dimensionality on its error convergence rate.\
n\nFurther\, we perform convergence analysis for two commonly used simulat
ion methods – stochastic mesh method and least squares method – to pri
ce American options in high dimensions. We derive respective optimal MSE c
onvergence rates for estimators based on the two methods and compare their
performance.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=3868
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3868
END:VEVENT
BEGIN:VEVENT
SUMMARY:Discriminative Pronunciation Modeling for Dialectal Speech Recogni
tion
DTSTART;VALUE=DATE-TIME:20140903T093000Z
DTEND;VALUE=DATE-TIME:20140903T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3875@cern.ch
DESCRIPTION:Speech recognizers are typically trained with data from a stan
dard dialect and do not generalize to non-standard dialects. Mismatch main
ly occurs in the acoustic realization of words\, which is represented by a
coustic models and pronunciation lexicon. Standard techniques for addressi
ng this mismatch are generative in nature and include acoustic model adapt
ation and expansion of lexicon with pronunciation variants\, both of which
have limited effectiveness. We present a discriminative pronunciation mod
el whose parameters are learned jointly with parameters from the language
models. We tease apart the gains from modeling the transitions of canonica
l phones\, the transduction from surface to canonical phones\, and the lan
guage model. We report experiments on African American Vernacular English
(AAVE) using NPR's StoryCorps corpus. Our models improve the performance o
ver the baseline by about 2.1% on AAVE\, of which 0.6% can be attributed t
o the pronunciation model. The model learns the most relevant phonetic tra
nsformations for AAVE speech.\n\nBio: Izhak Shafran is a speech researcher
\, who has been working on acoustic modeling and large vocabulary speech r
ecognition since 1996. Before joining Google\, he was an Associate Profess
or and a member of the Center for Spoken Language Processing at OHSU\, whe
re his focus was on medical application specifically on Parkinsons' diseas
e\, depression and mild congitive impairment. He graduated from University
of Washington in Seattle in 2001 and subsequently worked at AT\\&T Resear
ch Labs at Florham Park with the speech algorithms group. In summer of 200
6\, he was a visiting professor at Univeristy of Paris-South\, working at
LIMSI. Subsequently\, he was a research faculty at the Center for Language
and Speech Processing (CLSP) in Johns Hopkins University. He received an
NIH Career Development Award in 2010. He started his research career at Ta
ta Institute of Fundamental Research in Radio Astronomy.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=3875
LOCATION:Colaba Campus AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3875
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Classical Central Limit Theorem in Wasserstein Metric
DTSTART;VALUE=DATE-TIME:20140919T083000Z
DTEND;VALUE=DATE-TIME:20140919T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3908@cern.ch
DESCRIPTION:We discuss a proof of the classical CLT (in Wasserstein metric
) using Stein's Lemma.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=3908
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3908
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hardness of Vertex Guarding Polygons With Holes
DTSTART;VALUE=DATE-TIME:20140926T083000Z
DTEND;VALUE=DATE-TIME:20140926T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3916@cern.ch
DESCRIPTION:The VERTEX GUARD (VG) problem is defined as follows: Given a p
olygon P (with holes allowed) with n vertices\, find a smallest subset S o
f the set of vertices of P such that every point in the polygon P can be s
een from at least one vertex in S. The vertices in S are called the vertex
guards.\n\nIn this talk\, we will prove that VERTEX GUARD is NP-hard by s
howing a reduction from SET COVER which was put forward by Eidenbenz\, Sta
mm and Widmayer in 1998. If time permits\, we will also present our own mo
difications to this reduction that proves that VERTEX GUARD is NP-hard eve
n when considering the special class of polygons with holes that are weakl
y visible from an edge. In fact\, we can show that there cannot exist a po
lynomial time algorithm for the vertex guard problem with an approximation
ratio better than $((1−\\epsilon)/12)\\ln n$ for any $\\epsilon>0$\, un
less NP $\\subseteq$ TIME($n^{\\mathcal{O}(\\log\\log n)}$).\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=3916
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3916
END:VEVENT
BEGIN:VEVENT
SUMMARY:Busy Beavers and Numbers Left Untouched
DTSTART;VALUE=DATE-TIME:20141010T083000Z
DTEND;VALUE=DATE-TIME:20141010T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3941@cern.ch
DESCRIPTION:In the first half of this talk\, I will talk about the halting
problem and show that there are well defined functions f:N->N that no com
putable function can dominate. In the second half\, I will introduce Kolmo
gorov complexity and present Chaitin's Incompleteness theorem. If time per
mits\, I will also show a proof of Goodstein's theorem.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=3941
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3941
END:VEVENT
BEGIN:VEVENT
SUMMARY:Frobenius Automorphism and Quantum Cyclic Codes
DTSTART;VALUE=DATE-TIME:20141014T043000Z
DTEND;VALUE=DATE-TIME:20141014T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3939@cern.ch
DESCRIPTION:Efficient quantum codes are a necessary building block in the
eventual design of quantum computers. Despite some of the strange laws of
quantum physics that seems difficult to handle --- the no cloning theorem
for example which states one cannot make copies of quantum information ---
the theory of quantum codes\, in particular stabilize codes\, have a rema
rkable similarity to that of classical linear codes. The talk aims at givi
ng a general introduction to the theory of quantum codes followed by a dis
cussion on some of our recent work in this area. Starting with classical l
inear codes\, we quickly build up the necessary background on quantum stab
ilizer codes.\n\nHaving built the necessary background\, I describe\, in t
he second part of the talk\, our recent work on constructing families of c
yclic codes which we call the Frobenius codes. These codes include the wel
l known Laflamme [5\,1\,3] code. We give efficient decoding algorithms tha
t generalise the Berlekamp algorithm for classical cyclic codes. The theor
y we develop also proves that cyclic quantum codes for certain lengths do
not exists. These negative results for quantum codes are purely field theo
retic and not based on packing bounds (this is joint work with Sagarmoy Du
tta).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=393
9
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3939
END:VEVENT
BEGIN:VEVENT
SUMMARY:Light-trail and FISSION Architectures for Metropolitan and Data-Ce
nter Networking
DTSTART;VALUE=DATE-TIME:20141016T053000Z
DTEND;VALUE=DATE-TIME:20141016T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3926@cern.ch
DESCRIPTION:The talk begins by highlighting the communication problem in m
etropolitan regions and for content providers i.e. the use of data-centers
. The concept of light-trails is proposed as a solution to solving the IP
over WDM paradox. Light-trail implementation and protocols for growth of a
virtual topology are discussed. Subsequently\, implementation of light-tr
ails in commercial scenarios is highlighted. The second part of the talk u
ses light-trails as an enabler towards data-center design. The resuling ar
chitecture called FISSION or Flexible Interconnection of Scalable Systems
Integrated using Optical Networks is presented. FISSION allows a data-cent
er to scale to previously un-achievable number of servers. Engineering cha
llenges and optimization aspects of FISSION are presented. The talk conclu
des with new opportunities for research in the metropolitan and data-cente
r space.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
3926
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3926
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sub-logarithmic Approximation for Two Variations of tollbooth Prob
lem
DTSTART;VALUE=DATE-TIME:20141017T083000Z
DTEND;VALUE=DATE-TIME:20141017T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3955@cern.ch
DESCRIPTION:We consider the tollbooth pricing problem where the input is a
tree with n nodes and m paths\, also called customers\, with budgets b_i\
, i \\in [m]. Each customer wants to buy her path if the budget permits\,
otherwise she does not buy anything. The goal is to come up with a pricing
of the edges (non-negative in our cases) such that the revenue collected
is maximized. This problem is known as tollbooth pricing problem. Several
variants of this problem have appeared in literature\, all of them turn ou
t to be hard to approximate. Interestingly\, if we restrict the underlying
tree to a path\, the problem usually admits very good approximation (or i
s even poly-time solvable).\n\nIn this talk\, we will discuss two variants
of this problem\, namely\, uniform tollbooth problem and unique coverage
tollbooth problem. In uniform tollbooth problem\, the budgets of the custo
mers are within a constant factor of each other and in unique coverage pro
blem the budget of each customer as well as the pricing of each edges can
be either 0 or 1. In other words\, there is a set of edges X with the rest
riction that each customer can buy at most 1 edge from X. In recent work o
f Cygan et al. (ESA 2012)\, it is shown that both of these two problems ad
mit a \\mathcal{O}(\\log \\log n)-approximation algorithms.\n\nIn this tal
k we will further restrict the problem and show the results for trees with
small diameter. We will also mention how this technique can be extended f
or arbitrary trees.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=3955
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3955
END:VEVENT
BEGIN:VEVENT
SUMMARY:3-term Arithmetic Progression Free Sets
DTSTART;VALUE=DATE-TIME:20141024T083000Z
DTEND;VALUE=DATE-TIME:20141024T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3978@cern.ch
DESCRIPTION:A set S is said to be 3-term A.P free if there are no elements
in S which form a 3-term A.P.In first part of talk\, we will see a greedy
way of constructing a 3-term A.P free set S.In the later part\, we state
and prove Behrend's theorem which shows that there exists set A(subset of
[N]) which is 3-term free s.t |A|>>N.exp(-csqrt(logN)) for some c>0.\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3978
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3978
END:VEVENT
BEGIN:VEVENT
SUMMARY:Uncovering Causality
DTSTART;VALUE=DATE-TIME:20141031T083000Z
DTEND;VALUE=DATE-TIME:20141031T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3983@cern.ch
DESCRIPTION:Correlations\, as you know\, is not causation\; in fact\, a va
st majority of correlations aren't causal. However\, human progress has ha
ppened not because someone understood that the banging of a hammer was cor
related with the burial of a nail but because someone understood that the
banging of a hammer caused the burial of the nail\; we are interested in d
oing things and having those actions cause desirable consequences.\n\nEarl
ier\, the only way to abstract causation from data was randomised controll
ed trials - and those are often broken by confounders. In the past couple
of decades\, Judea Pearl and his collaborators have begun uncovering causa
l relationships as properties of particular conditionalisations of joint p
robability distributions - allowing us to infer causal relationships from
data not found using RCTs. And as a bonus\, this method of analysis is als
o robust to identified confounders.\n\nI'll be discussing the general idea
s of this analysis\, and maybe even some simple examples.\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=3983
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3983
END:VEVENT
BEGIN:VEVENT
SUMMARY:Universal Convexification via Risk Aversion
DTSTART;VALUE=DATE-TIME:20141107T090000Z
DTEND;VALUE=DATE-TIME:20141107T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-3993@cern.ch
DESCRIPTION:We develop a framework for convexifying a fairly general class
of optimization problems. Under additional assumptions\, we analyze the s
uboptimality of the solution to the convexified problem relative to the or
iginal nonconvex problem and prove additive approximation guarantees. We t
hen develop algorithms based on stochastic gradient methods to solve the r
esulting optimization problems and show bounds on convergence rates. We th
en extend this framework to apply to a general class of discrete-time dyna
mical systems. In this context\, our convexification approach falls under
the well-studied paradigm of risk-sensitive Markov Decision Processes. We
derive the first known model-based and model-free policy gradient optimiza
tion algorithms with guaranteed convergence to the optimal solution. Final
ly\, we present numerical results validating our formulation in different
applications.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=3993
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=3993
END:VEVENT
BEGIN:VEVENT
SUMMARY:Complete Characterization of Two-party Securely Computable Functio
ns
DTSTART;VALUE=DATE-TIME:20141114T083000Z
DTEND;VALUE=DATE-TIME:20141114T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4012@cern.ch
DESCRIPTION:In two-party secure computation\, Alice has an input X\, Bob h
as an input Y and both of them want to compute a function $f:\\mathcal{X}
\\times \\mathcal{Y} \\to \\mathcal{Z}$ securely by exchanging messages to
each other over several rounds. By security we mean that they should comp
ute $f(X\,Y)$ correctly and either party should not learn any information
about the other party's input other than what it can infer from its own in
put and the function value. We will characterize the class of securely com
putable functions\, i.e.\, which functions can be computed securely and wh
ich cannot be.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=4012
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4012
END:VEVENT
BEGIN:VEVENT
SUMMARY:Compression with Uncertain Priors Using Imperfectly Shared Randomn
ess
DTSTART;VALUE=DATE-TIME:20141121T083000Z
DTEND;VALUE=DATE-TIME:20141121T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4032@cern.ch
DESCRIPTION:In this talk\, we will go through some of the results in the p
aper "Communication with Imperfectly Shared Randomness by Canonne\, Gurusw
ami\, Meka and Sudan.\n\nThe problem of compression with uncertain priors
is: Alice has a distribution P on a universe [N] = {1\,...\,N}\, and a mes
sage m ∈ [N] chosen according to the distribution P. Alice is allowed to
send some bits to Bob and Bob should output m and the goal is to minimize
the expected number of bits that Alice sends Bob (over the random choice
of m). If Bob knows the distribution P exactly then this is the classical
compression problem\, solved for example by Huffman coding. In most forms
of natural communication (e.g.\, think about the next email you are about
to send)\, Alice and Bob are not perfectly aware of the underlying context
to their exchange\, but have reasonably good ideas about each other. One
way to model this is to say that Bob has a distribution Q that is close to
the distribution P that Alice is working with\, but is not identical to P
. Compressing information down to its entropy in the presence of such unce
rtainty (i.e.\, P 6= Q) turns out to be possible if Alice and Bob share ra
ndomness that is independent of (P\,Q\,m) as shown by Juba et al. . It is
natural to ask the question: can the (presumed) savings in communication b
e achieved in the absence of perfect sharing of randomness?\n\nIf time per
mits we will also look at other problems where solutions with perfect shar
ing of randomness can be extended to the imperfect setting.\n\nhttps://ind
ico.tifr.res.in/indico/conferenceDisplay.py?confId=4032
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4032
END:VEVENT
BEGIN:VEVENT
SUMMARY:Unambiguity and Timed Languages : Automata\, Logics\, Expressivene
ss
DTSTART;VALUE=DATE-TIME:20141202T053000Z
DTEND;VALUE=DATE-TIME:20141202T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4044@cern.ch
DESCRIPTION:Regular languages exhibit considerable structure. The characte
rization of various classes of regular languages using classical and tempo
ral logics\, automata and varieties of monoids\, as an attempt towards uni
fication of diverse frameworks\, has been an important theme and area of r
esearch for many decades. In this thesis\, we have investigated logic-auto
mata connections for unambiguous languages over finite words and also time
d words.\n\nUnambiguous star-free regular languages (UL) originally introd
uced by Schutzenberger\, have various characterizations within classical l
ogics (FO2[\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=4044
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4044
END:VEVENT
BEGIN:VEVENT
SUMMARY:Information Theory for DNA / RNA Sequencing
DTSTART;VALUE=DATE-TIME:20141205T053000Z
DTEND;VALUE=DATE-TIME:20141205T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4042@cern.ch
DESCRIPTION:Extraordinary advances in sequencing technology in the past de
cade have revolutionized biology and medicine. In this talk\, we will surv
ey some of our work on an information theoretic approach to some string re
construction problems that arise from RNA sequencing and for DNA variant c
alling. Algorithms derived from this approach are provably near-optimal in
the information theoretic sense\, while simultaneously being computationa
lly efficient. Interesting connections to sparse flow decomposition in net
works and to iterative algorithms will be discussed. To highlight the bene
fits of this approach\, we will also describe our development of new softw
are package that achieves significant informational gains over existing so
ftware for RNA assembly. This software has applications in several areas o
f biology and medicine including cancer.\n\nBio: Sreeram Kannan is current
ly a faculty at University of Washington\, Seattle. He was a postdoctoral
scholar at University of California\, Berkeley till recently\, before whic
h he received his Ph.D. in Electrical Engineering and M.S. in mathematics
from the University of Illinois Urbana Champaign. He is a recipient of the
Van Valkenburg research award from UIUC\, 2013\, a co-recipient of the Qu
alcomm Cognitive Radio Contest first prize\, 2010\, a recipient of Qualcom
m CTO Roberto Padovani outstanding intern award\, 2010\, a recipient of th
e S.V.C. Aiya medal from the Indian Institute of Science\, 2008\, and a co
-recipient of Intel India Student Research Contest first prize\, 2006. His
research interests include the applications of information theory and app
roximation algorithms to computational biology and wireless networks.\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4042
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4042
END:VEVENT
BEGIN:VEVENT
SUMMARY:Size-sensitive Packing Number for the Hamming Cube and its Consequ
ences
DTSTART;VALUE=DATE-TIME:20141205T083000Z
DTEND;VALUE=DATE-TIME:20141205T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4058@cern.ch
DESCRIPTION:An abstract set system\, or hypergraph\, consists of a univers
e (here assumed finite)\, together with a subset of its power set. A set-s
ystem is $\\delta$-separated if any pair of its constituent subsets have s
ymmetric difference at least $\\delta$. In 1992\, Haussler proved an optim
al upper bound on the packing number for $\\delta$-separated set systems h
aving bounded primal shatter dimension. A set system with bounded primal s
hatter dimension is said to have \\emph{size-sensitive shattering constant
s} $d_1$ and $d_2$\, if the number of distinct projections of the system o
n any subset of its universe having $m$ elements\, is at most $O(m^{d_1}k^
{d_2})$.\n\nWe prove a size-sensitive version of Haussler’s Packing lemm
a [Hau92] for set-systems with bounded primal shatter dimension\, which ha
ve an additional size-sensitive property. This answers a question asked by
Ezra [Ezr14]. As a consequence of this result we get an improvement on th
e discrepancy bounds for set systems with the above size sensitive propert
y. Improved bound on the discrepancy for these special set systems also im
plies an improvement in the size of $(\\nu\, \\alpha)$-samples (and relati
ve $(\\vareps\,\\delta)$-approximations) (joint work with Arijit Ghosh\, M
ax-Planck-Institute\, Saarbrücken).\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=4058
LOCATION: AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4058
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Attempt to Derandomize the Random Unitary Matrices Used for Dec
oupling in Quantum Information Theory
DTSTART;VALUE=DATE-TIME:20141208T063000Z
DTEND;VALUE=DATE-TIME:20141208T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4066@cern.ch
DESCRIPTION:The correlation of a quantum system $A$ to an initially correl
ated system $E$ generally decreases when $A$ undergoes an evolution separa
ted from $E$. Here we study the conditions under which the correlation dis
appears completely\, resulting in a decoupling of $A$ from $E$. Decoupling
is a general way to prove many coding theorems in quantum information the
ory. However existing decoupling constructions use random unitaries which
are inefficient to implement. Hence the quest for derandomization.\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4066
LOCATION: D-406 (D-Block Conference Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4066
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quest for an Explicit Quantum Channel with Super-additive Classica
l Capacity
DTSTART;VALUE=DATE-TIME:20141208T053000Z
DTEND;VALUE=DATE-TIME:20141208T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4072@cern.ch
DESCRIPTION:Study of super operators\, norms on super operators (referred
to as super operator norms) and the minimum output -Re'nyi entropy of a ty
pical quantum channel. These are used to study the additivity conjecture a
nd the argument used by Hayden–Winter disproving the additivity conjectu
re for the minimal output p-Re'nyi entropy (for > 1).\n\nhttps://indico.ti
fr.res.in/indico/conferenceDisplay.py?confId=4072
LOCATION: D-406 (D-Block Conference Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4072
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mobile Sensing: Efficient Sampling and Privacy Concerns
DTSTART;VALUE=DATE-TIME:20141215T050000Z
DTEND;VALUE=DATE-TIME:20141215T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4079@cern.ch
DESCRIPTION:Sensing of spatial fields is traditionally studied in a settin
g where static sensors take measurements of the spatial field at their loc
ations. However many modern applications like citizen sensing and robotic
sensing employ moving sensors for spatial sensing. This emerging paradigm
of mobile sensing requires us to rethink the classical notions of sampling
and privacy. In this talk I will focus on the following aspects.\n\n(1) A
sampling theory for mobile sensing: We introduce the notion of path densi
ty\, defined as the total distance traveled by the mobile sensors per unit
spatial volume. We design sensor trajectories that are efficient in terms
of path density for sampling spatially bandlimited fields\, and obtain fu
ndamental limits on the path density of mobile sensor trajectories that ad
mit stable sampling. These limits are analogous to Landau-Nyquist rates fo
r classical sampling (joint work with M. Vetterli\, J.L. Romero\, K. Groch
enig).\n\n(2) Privacy of mobility statistics: How private are anonymized m
obility statistics? We study the de-anonymization of anonymized mobility s
tatistics as a hypothesis testing problem\, and identify the optimal schem
e for matching anonymized location histograms to auxiliary observations of
the users' locations. We apply the scheme to datasets of Wi-Fi traces\, c
all data records\, and web browsing histories to highlight the privacy con
cerns in collecting location information in citizen sensing schemes (joint
work with F. Naini\, P. Thiran\, M. Vetterli).\n\nIn sum\, citizen sensin
g opens up possibilities for efficient spatial sampling at the risk of rai
sing privacy issues. I will quantify both aspects in this talk and present
new algorithms and demonstrate their usefulness by applying them on real
data.\n\nBio: Jayakrishnan Unnikrishnan received the B.Tech. degree in ele
ctrical engineering from the Indian Institute of Technology\, Madras in 20
05 and the M.S. and Ph.D. degrees in electrical and computer engineering f
rom the University of Illinois at Urbana-Champaign in 2007 and 2010\, resp
ectively.\n\nFrom 2010 to 2014\, he worked as a postdoctoral researcher at
the School of Computer and Communication Sciences\, Ecole Polytechnique F
ederale de Lausanne (EPFL)\, Lausanne\, Switzerland. His current research
interests include signal processing\, detection and estimation theory\, an
d information theory.\n\nDr. Unnikrishnan is a recipient of the Vodafone G
raduate Fellowship Award from the University of Illinois at Urbana-Champai
gn for 2007--2008 and the E.A. Reid Fellowship Award from the ECE departme
nt at the University of Illinois at Urbana-Champaign for 2010--2011.\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4079
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4079
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rare events in Heavy-tailed Stochastic Systems: Algorithms and Ana
lysis
DTSTART;VALUE=DATE-TIME:20141216T090000Z
DTEND;VALUE=DATE-TIME:20141216T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4091@cern.ch
DESCRIPTION:Events that occur with low probability are called rare events.
In the design of probabilistic models\, it is important to ensure that ce
rtain undesirable events\, for example\, bankruptcy of financial instituti
ons (or) data loss in communication networks\, do not happen (or) if at al
l they happen\, they happen with low probability. Conventional theory of l
arge deviations\, which is usually used to explain the behaviour of such r
are events\, is inapplicable when the random variables involved are heavy-
tailed. We call a random variable to be heavy-tailed if its tail probabili
ties decay slowly at some polynomial rate. Simulation of rare events\, in
particular\, has been traditionally considered difficult when heavy-tailed
random variables are involved. In this work\, we analyse asymptotic behav
iour of certain rare events involving heavy-tailed random variables and de
velop simulation algorithms that are provably efficient. To be specific\,
we discuss the following contributions in the talk:\n \n1) Based on the "b
ig jump principle"\, we develop an entirely new methodology for the estima
tion of various large deviations and level crossing probabilities that ari
se in the context of heavy-tailed sums. Our key contribution has been to q
uestion the prevailing view that one needs to resort to state-dependent me
thods when a large number of heavy-tailed random variables are involved. T
he algorithms we develop are provably efficient\, follow a general templat
e\, and are easy to generalize\, as we shall see\, to settings more genera
l than sums of independent random variables. In addition to simulation inv
olving finite sums\, we tackle the problem of estimating tail probabilitie
s of sums of infinite sums\, where bias is generally a problem. Apart from
eliminating bias\, our algorithm computes tail probabilities with uniform
ly bounded computational effort.\n \n2) Building up on the theory of rare
events in heavy-tailed sums\, we attempt to explain how large delays in se
rvice happen in multi-server queues when incoming jobs are heavy-tailed (i
n size) and traffic intensity is an integer. In the current literature\, c
haracterizations of steady-state delay exist only when the traffic intensi
ty is not an integer. There are qualitative reasons\, as we shall see\, th
at makes the integer case more delicate to analyse. Our main contribution
is that we develop the first known tail asymptotic for steady-state delay
in multi-server queues when the traffic intensity is an integer. Specifica
lly\, we consider a two-server queue and identify interesting transitions
in the tail behaviour of steady-state delay via a careful analysis that is
not typical in the analysis of queuing systems.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=4091
LOCATION:
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4091
END:VEVENT
BEGIN:VEVENT
SUMMARY:Efficient Detection and Localization on Graph Structured Data
DTSTART;VALUE=DATE-TIME:20141219T090000Z
DTEND;VALUE=DATE-TIME:20141219T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4094@cern.ch
DESCRIPTION:We consider stochastic sequential learning problems where the
learner can observe the average reward of several actions. Such a setting
is interesting in many applications involving monitoring and surveillance\
, where the set of the actions to observe represent some (geographical) ar
ea. The importance of this setting is that in these applications\, it is a
ctually cheaper to observe average reward of a group of actions rather tha
n the reward of a single action. We show that when the reward is smooth ov
er a given graph representing the neighboring actions\, we can maximize th
e cumulative reward of learning while minimizing the sensing cost. We prop
ose CheapUCB\, an algorithm that matches the regret guarantees of the know
n algorithms for this setting and at the same time guarantees a linear cos
t again over them.\n\nBio: Manjesh Kumar Hanawal received Ph.D. degree at
INRIA\, Sophia Antipolis\, France\, and the University d’Avignon\, Avign
on\, France in 2013. Before this he received B.E. degree in electronics an
d communication from the National Institute of Technology\, Bhopal\, India
\, and the M.Sc. degree in electrical communication engineering from the
Indian Institute of Science\, Bangalore\, India. Currently\, he is a Postd
octoral associate at Boston University\, USA. His bachelor's studies was s
ponsored by Airtel India selecting him as a Bharti scholar. His masters’
thesis was awarded best M.Sc. thesis medal (Prof. F. M. Mowdawalla Medal)
\, and he is recipient of INSPIRE faculty fellowship from DST\, Government
of India.\n\nFrom 2004 to 2007\, he was with the Centre for Artificial In
telligence and Robotics (CAIR)\, DRDO\, Bangalore\, as Scientist-B.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4094
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4094
END:VEVENT
BEGIN:VEVENT
SUMMARY:Load Balancing in Large Graphs
DTSTART;VALUE=DATE-TIME:20141222T103000Z
DTEND;VALUE=DATE-TIME:20141222T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4098@cern.ch
DESCRIPTION:We consider load balancing on a large graph. Each edge has a u
nit of load that it wishes to distribute between its nodes in the most bal
anced way. For infinite graphs the corresponding load balancing problem ex
hibits nonuniqueness\, related to role of boundary conditions in statistic
al mechanical models.\n\nNevertheless\, we are able to extend the notion o
f balanced loads from large finite graphs to their local weak limits\, usi
ng the concept of unimodularity. The result applies in particular to the E
rd"os-R'enyi model\, where it settles a conjecture of Hajek (1990). Our p
roof is a new illustration of the objective method described by Aldous and
Steele (2004).\n\nAll the necessary background from the machinery of loca
l weak convergence that is needed will be developed during the talk. This
machinery provides a way to study many problems of applied interest in lar
ge networks beyond just the load balancing problem that will be the focus
of this talk. It is therefore valuable to familiarize oneself with this th
eory if one is interested in understanding the behavior of large networks
such as wired and wireless networks\, transportation networks\, and social
networks (joint work with Justin Salez\, Universit'e Paris Diderot).\n\nB
io: Venkat Anantharam is on the faculty of the EECS Department at the Univ
ersity of California\, Berkeley. He received the Philips India Medal and t
he President of India Gold Medal from IIT Madras in 1980 and an NSF Presid
ential Young Investigator award in 1988. He is a corecipient of the 1998 P
rize Paper Award of the IEEE Information Theory Society\, and a corecipien
t of the 2000 Stephen O. Rice Prize Paper Award of the IEEE Communications
Theory Society. He received the Distinguished Alumnus Award from IIT Madr
as in 2008.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=4098
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4098
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Tale of Two Measures
DTSTART;VALUE=DATE-TIME:20150102T090000Z
DTEND;VALUE=DATE-TIME:20150102T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4101@cern.ch
DESCRIPTION:Information theory has been traditionally studied in the conte
xt of communication theory and statistical physics. However\, it has also
had important applications in other fields such as computer science\, econ
omics\, mathematics\, and statistics. This talk is very much in the spirit
of discovering applications of information theory in other fields. We wil
l discuss three such recent applications:\n\nStatistics: The Hirschfeld-Ge
belein-Rényi maximal correlation is an important tool in statistics that
has found numerous applications from correspondence analysis\, to detectio
n of non-linear patterns in data. We will describe a simple information-th
eoretic proof of a fundamental result on maximal correlation due to Dembo\
, Kagan\, and Shepp (2001).\n\nComputer Science: Boolean functions are one
of the most basic objects of study in theoretical computer science. We sh
ow how information-theoretic tools can aid Fourier analytic tools in this
quest. Specifically\, we will consider the problem of correlation between
Boolean functions on a noisy hypercube graph.\n\nMathematics: Hypercontrac
tivity and Reverse Hypercontractivity are very useful tools for studying c
oncentration of measure\, and extremal questions in the geometry of high-d
imensional spaces\, both discrete and continuous. In this talk\, we will d
escribe a recent result by Chandra Nair characterizing hypercontractivity
using information measures. We will extend this result to reverse hypercon
tractivity\, and we will discuss implications of these results.\n\nThe tit
le of this presentation is derived from two measures of correlation\n\n- t
he maximal correlation and the so-called strong data processing constant\n
- that will be key concepts used throughout.\n\nThis talk is based on join
t work with Venkat Anantharam\, Amin Gohari\, and Chandra Nair.\n\nhttps:/
/indico.tifr.res.in/indico/conferenceDisplay.py?confId=4101
LOCATION:
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4101
END:VEVENT
BEGIN:VEVENT
SUMMARY:Algorithmic Problems in Higher-order Fourier Analysis
DTSTART;VALUE=DATE-TIME:20150103T043000Z
DTEND;VALUE=DATE-TIME:20150103T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4120@cern.ch
DESCRIPTION:Decomposition theorems proved by Gowers and Wolf provide an ap
propriate notion of "Fourier transform" for higher-order Fourier analysis.
We will discuss some questions and techniques that arise from trying to d
evelop polynomial time algorithms for computing these decompositions.\n\nT
he original proofs for these theorems were non-constructive and used the H
ahn-Banach theorem. We will discuss constructive proofs based on\nboosting
which reduce the problem of computing these decompositions to a certain k
ind of weak decoding for codes beyond the list-decoding radius. We will al
so describe some special cases for which such decodings are known to be po
ssible\, and the techniques which achieve these.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=4120
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4120
END:VEVENT
BEGIN:VEVENT
SUMMARY:One-shot Bounds in Classical and Quantum Information Theory
DTSTART;VALUE=DATE-TIME:20150110T053000Z
DTEND;VALUE=DATE-TIME:20150110T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4127@cern.ch
DESCRIPTION:The goal of information theory is to understand the limits of
data compression and communication in the presence of noise. Traditionall
y in information theory literature it is common to study the underlying pr
oblems in the asymptotic setting\, often assuming that the channel charact
eristics do not change over multiple use. The proofs appeal to typicality
of sequences or subspaces: the empirical distribution of symbols in a long
sequence of trials will with high probability be close to the true distri
bution. However\, information theoretic arguments based on typicality ass
ume that both the source and channel are stationary and/or ergodic (memory
less)\, assumptions that are not always valid.\n\nIn this thesis we study
some information theoretic protocol in the one-shot setting where we are a
llowed to use the channel only once and are allowed only a small probabili
ty of error. Such results are more general\, for one can always view a cha
nnel used repeatedly as a single channel with a larger alphabet\, and reco
ver the asymptotic bounds as a special case. The general one-shot results\
, while sometimes technically harder to show (for arguments based on typic
ality is no longer available)\, are often so strong that nothing is lost i
n deriving the asymptotic results from them. We discuss the following con
tributions in the talk:\n\n1) One-shot bounds on the coding rates for sour
ce coding of correlated sources.\n\n2) One-shot bounds on the reliability
of information transmission over the quantum channel.\n\n3) One-shot Marto
n inner bound for the classical-quantum broadcast channel.\n\n4) One-shot
bounds on the private capacity of the quantum channel.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=4127
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4127
END:VEVENT
BEGIN:VEVENT
SUMMARY:Communication with Imperfectly Shared Randomness
DTSTART;VALUE=DATE-TIME:20150112T090000Z
DTEND;VALUE=DATE-TIME:20150112T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4132@cern.ch
DESCRIPTION:In the 1980's Yao introduced the model of communication comple
xity where Alice\, who knows x in 0\,1n\, and Bob\, who knows y in 0\,1n\,
wish to communicate with each other\, exchanging fewest possible number o
f bits\, to determine some function f(x\,y). For many natural functions f
this communication complexity is much smaller than the trivial n bits sugg
esting that communication with a goal in mind may be much less expensive t
han otherwise. This message gets amplified even more if Alice and Bob shar
e some random string r chosen independently of x and y: in this setting ev
en more functions f can be determined quickly (with high probability).\n\n
In this talk I will introduce a relaxation of this model where Alice and B
ob do not share the random string perfectly\, but rather Alice knows r and
Bob knows some string s that is correlated with r. I will describe some r
ecent results showing that any communication protocol with k bits of commu
nication between Alice and Bob with perfect sharing of randomness\, contin
ues to have a moderately low-complexity\, 2k bit\, protocol with shared co
rrelation. Furthermore\, I will show that this result is tight in that the
re exist problems where this exponential jump is necessary. The technical
core of these results rely on the understanding of the influence of variab
les in the analysis of Boolean functions and recently developed probabilis
tic tools such as the ``invariance principle'' of Mossel\, O'Donnell and O
leszkiewicz. The talk will assume no background in communication complexit
y and/or influence of variables.\n\nBased on joint work with Clément Cano
nne (Columbia)\, Venkatesan Guruswami (CMU) and Raghu Meka (UCLA).\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4132
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4132
END:VEVENT
BEGIN:VEVENT
SUMMARY:Early Verification
DTSTART;VALUE=DATE-TIME:20150113T083000Z
DTEND;VALUE=DATE-TIME:20150113T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4134@cern.ch
DESCRIPTION:echnology that accurately models\, analyzes\, and verifies sof
tware has come a long way since its conception several decades ago. One m
ode of using such technology is to look for defects in software that has a
lready left the hands of developers. Another mode is to integrate the tec
hnology into the process of software authoring. The advantage of this mod
e is that it lends analytical power to the developer's thinking. To be us
ed in this way\, the technology must be packaged in a way that is understa
ndable\, unobtrusive\, and responsive.\n\nIn this talk\, I showcase an int
egrated development environment that supports reasoning and verification\,
trying to provide an aid to the developer earlier during the software dev
elopment process.\n\nBio: Prof. Dr. K. Rustan M. Leino is a Principal Rese
archer in the Research in Software Engineering (RiSE) group at Microsoft R
esearch and a Visiting Professor in the Computing Department at Imperial C
ollege London. He is known for his work on programming methods\, language
s\, and verification\, and is a world leader in building automated program
verification tools. Languages and tools he has worked on include Dafny\,
Chalice\, Jennisys\, Spec#\, Boogie\, Houdini\, ESC/Java\, and ESC/Modula
-3. In the past\, he has worked as a researcher at DEC and Compaq\, and a
s a software engineer at Microsoft. He received his PhD in Computer Scien
ce from Caltech in 1995.\n\nLeino also collects thinking puzzles on a popu
lar web page and hosts the Verification Corner channel on youtube. In his
spare time\, he enjoys playing music\, cooking food\, and\, having ended
a 7-year tenure as group cardio exercise instructor\, learning to dance.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4134
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4134
END:VEVENT
BEGIN:VEVENT
SUMMARY:Information Geometry of the Robust Maximum Mean Power Likelihood E
stimation
DTSTART;VALUE=DATE-TIME:20150120T081500Z
DTEND;VALUE=DATE-TIME:20150120T091500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4138@cern.ch
DESCRIPTION:The maximum "mean power" likelihood estimation procedure (prop
osed by [Basu et al.\, Biometrika\, 1998]) is a robust variant of the maxi
mum likelihood estimation procedure. Given iid samples coming from an unkn
own distribution\, a member of a given parametric family of distributions\
, the mean power likelihood estimation procedure is to find the parameter
of the distribution that is closest\, in a particular sense\, to the empir
ical distribution of the samples. One may view the resulting distribution
as a "projection" of the empirical distribution on the parametric family.
In this talk\, I will highlight the geometry associated with this projecti
on. I will also discuss a simplified computation procedure\, one that is s
uggested by the geometric view point\, when the estimation is of the param
eter of a power-law family (this is joint work with M. Ashok Kumar). arXiv
:1410.5550\n\nBiography: Rajesh Sundaresan is an associate professor at th
e ECE department of IISc Bangalore.\n\nhttps://indico.tifr.res.in/indico/c
onferenceDisplay.py?confId=4138
LOCATION: AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4138
END:VEVENT
BEGIN:VEVENT
SUMMARY:Distance Preserving Minors in Graphs
DTSTART;VALUE=DATE-TIME:20150120T103000Z
DTEND;VALUE=DATE-TIME:20150120T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4148@cern.ch
DESCRIPTION:We consider undirected\, connected graphs with nonnegative wei
ghts on the edges. Additionally\, a special subset of vertices called term
inals is provided as input. A graph H is said to be a distance preserving
minor of G if: (i) H is a minor of G\, and (ii) the distance between each
pair of terminals is exactly the same in G and H. Note that the edge weigh
ts can be reassigned in H (as long as they are nonnegative). Given a famil
y of graphs F\, let f(k\,F) be the minimum integer such that every graph i
n F with k terminals admits a distance preserving minor with at most f(k\,
F) vertices. We will see results for the best known bounds on the value of
f for different families of graphs\, namely trees\, planar graphs and int
erval graphs.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=4148
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4148
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computational Challenges in Network Biology: From Brain Tissues to
Single Cells
DTSTART;VALUE=DATE-TIME:20150205T053000Z
DTEND;VALUE=DATE-TIME:20150205T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4169@cern.ch
DESCRIPTION:Interactions among proteins\, regulatory DNA regions surroundi
ng genes\, and other biomolecules are central to cellular function and hea
lth. Constructing a network representation of these interactions makes it
possible to dissect complex cellular behaviors using computational and sta
tistical methods. I would like to talk about the design and application of
a graph-theoretic algorithm that aligns gene networks derived from more t
han 600 human postmortem brain tissues to extract insights about dysregula
tion in two neurodegenerative diseases. Given the recent advances and huge
interest in single-cell measurement technologies\, I would like to conclu
de the talk with challenges in extending similar network inference ideas f
rom tissue-level measurements to those obtained from individual cells (thi
s work was supported by the Intramural Research Program of NIAID\, NIH).\n
\nBio: Manikandan Narayanan is a Staff Scientist in the Laboratory of Syst
ems Biology at the National Institutes of Health (NIH). He is interested i
n the design and application of computational methods to study cellular be
haviors. Complex cellular behaviors arise from interactions among various
molecules\, and Narayanan's specific interest involves analysis of network
(graph) representation of such interactions. He obtained his Ph.D. in Com
puter Science (with an emphasis in computational and genomic biology) from
the University of California at Berkeley under the mentorship of Prof. Ri
chard Karp\, and held a Sr. Research Scientist position at Merck Research
Labs first in Seattle and then in Boston before coming to the NIH. He is a
Siebel Scholar Class of 2003\, and has authored widely-cited publications
in top-ranking computational biology journals.\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=4169
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4169
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimality of Separation for Universal Communication With a Fideli
ty Criterion
DTSTART;VALUE=DATE-TIME:20150206T110000Z
DTEND;VALUE=DATE-TIME:20150206T123000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4202@cern.ch
DESCRIPTION:Optimality of source-channel separation for universal communic
ation with a fidelity criterion will be discussed. Universality is over th
e channel\; that is\, the channel can belong to a set.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=4202
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4202
END:VEVENT
BEGIN:VEVENT
SUMMARY:Partitioning Problems\, Submodular Functions and Approximating the
OFDMA Capacity
DTSTART;VALUE=DATE-TIME:20150210T103000Z
DTEND;VALUE=DATE-TIME:20150210T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4201@cern.ch
DESCRIPTION:Allocating disjoint resources to multiple users so as to maxim
ize the sum-utility is a classically hard problem. Finding the FDMA capaci
ty is one such special case\, where we have to allocate disjoint frequency
bins to multiple users so as to maximize the sum of the data rate. Typica
lly\, in wireless applications\, heuristic approaches such as convex relax
ation\, KKT conditions etc\,. are used to solve partitioning problem. We a
pproach this problem via results on greedy algorithms for sub-modular func
tions. The main result is to show that the capacity of parallel Gaussian c
hannels is a sub-modular function. Following this result\, we get a $2$-ap
proximation (which is at least 1/2 times equal) to the optimal FDMA capaci
ty. In addition\, exploiting the sub-modularity of the underlying utilitie
s\, we also derive a truthful mechanism where no user has any incentive of
misreporting its utility and still guarantee a price of anarchy of at mos
t 2.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4201
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4201
END:VEVENT
BEGIN:VEVENT
SUMMARY:Minmax Representation of Viscosity Solutions to Hamilton-Jacobi Eq
uations and Applications in Rare-event Simulation
DTSTART;VALUE=DATE-TIME:20150211T090000Z
DTEND;VALUE=DATE-TIME:20150211T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4177@cern.ch
DESCRIPTION:In this talk a duality relation between the Mane's potential a
nd Mather's action functional is derived in the context of convex and stat
e-dependent Hamiltonians. The duality relation is used to obtain min-max r
epresentations of viscosity solutions of first order Hamilton-Jacobi equat
ions. These min-max representations naturally suggest classes of subsolut
ions of Hamilton-Jacobi equations that arise in the theory of large deviat
ions. The subsolutions\, in turn\, are good candidates for designing effic
ient rare-event simulation algorithms. I will show some applications to fi
nancial risk management and reliability of power systems.\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=4177
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4177
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximating and Testing Equilibria
DTSTART;VALUE=DATE-TIME:20150212T090000Z
DTEND;VALUE=DATE-TIME:20150212T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4190@cern.ch
DESCRIPTION:Algorithmic game theory is an active and impactful area of res
earch that in recent years has found many applications in the study of lar
ge strategic environments\, like online markets and auctions\, as well as
social and biological systems. A basic message that emerged from this area
is in fact negative: determining game-theoretic solution concepts---e.g.\
, Nash equilibria---is computationally hard in general. But\, given that s
uch solution concepts are widely used in the design and analysis of strate
gic environments\, these complexity barriers need to be addressed.\n\nMoti
vated by these considerations\, in this talk I will present two complement
ary directions that address hardness results for Nash equilibria\, which a
re one of the most well-studied solution concepts in game theory. First\,
I will show that for a relevant class of two-player games an approximate N
ash equilibrium can be efficiently determined. A key technical component o
f this work is an approximate version of Caratheodory’s theorem\; given
the fundamental importance of Caratheodory’s theorem\, this approximate
version is interesting in its own right. Then\, I will describe how an emp
irical perspective can be employed to address the complexity of Nash equil
ibria in a number of cases. I will conclude by presenting a number of rela
ted problems and future research directions.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=4190
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4190
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fast and Efficient Algorithms for Sparse Recovery: A General Frame
work
DTSTART;VALUE=DATE-TIME:20150219T060000Z
DTEND;VALUE=DATE-TIME:20150219T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4215@cern.ch
DESCRIPTION:Sparse Recovery refers to a broad collection of techniques tha
t aim to exploit sparsity to dramatically reduce the "cost" of reconstruct
ing a low dimensional "sparse" dataset embedded in a prohibitively high di
mensional space. While sparse recovery techniques have been around for a l
ong time\, the seminal works on Compressive Sensing by Candes\, Tao\, and
Donoho\, have renewed interest in this area -- especially among informatio
n theorists. Formally\, consider an unknown "k-sparse vector" 'x'\, i.e.\
, a 'n'-length vector with at most 'k' non-zero coordinates. A typical Spa
rse Recovery problem involves performing m measurements on 'x' to obtain a
vector 'y=A(x)'. The goal is to be able to minimize 'm' while being able
to reconstruct 'x' from 'y' efficiently. The function 'A(x)' may be linea
r (e.g. compressive sensing) or non-linear (e.g. phase retrieval).\n\nIn t
his talk\, we will discuss a new framework -- "picking and peeling" -- for
fast and efficient algorithms for four important sparse recovery problems
-- compressive sensing\, compressive phase retrieval\, group testing\, an
d network tomography. We will begin the exposition through a small puzzle
and use it to explain the key ideas of picking and peeling. Building upon
this insight\, we will next describe our compressive sensing algorithm SHO
-FA (for SHOrt and FAst) which achieves a decoding complexity of O(k) whil
e using only O(k) measurements (this is the fastest possible performance i
n the order sense). Next\, we will briefly describe our algorithms for the
other three problems. For each of these problems\, our algorithms are eit
her order-optimal or near-optimal both in terms of two metrics - the numbe
r of measurements and the time complexity of decoding. Finally\, we will e
xamine another metric for performance - the energy required by the circuit
that implements these algorithms. Surprisingly\, even algorithms that are
efficient in terms of time complexity are no longer efficient in terms of
decoding energy. We derive a novel lower bound on the decoding energy req
uired in compressive sensing and outline a new algorithm that matches this
bound upto a constant factor (this talk is based on joint works with Sidh
arth Jaggi\, Minghua Chen\, Pulkit Grover\, Sheng Cai\, Tongxin Li\, and M
ohammad Jahangoshahi).\n\nBio: Prof Mayank Bakshi is affiliated with the I
nstitute of Network Coding (INC) at the Chinese University of Hong Kong. H
e finished his PhD from California Institute of Technology in 2011 and was
a Postdoctoral Scholar at INC from 2012 to 2014. Prior to this\, he did h
is B. Tech and M.Tech from IIT Kanpur in 2003 and 2005 respectively. He is
interested in a variety of Information Theoretic problems including Spars
e Recovery\, Secure Network Coding\, and Network Data Compression.\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4215
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4215
END:VEVENT
BEGIN:VEVENT
SUMMARY:Studying Lower Bounds for Conflict-free Colouring of Guards in Pol
ygons
DTSTART;VALUE=DATE-TIME:20150219T103000Z
DTEND;VALUE=DATE-TIME:20150219T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4224@cern.ch
DESCRIPTION:Art gallery problem is to determine the number of guards that
are sufficient to see or cover all points in $P$. Consider the problem of
placing colour guards in an $n$-sided polygon $P$ such that every point $z
\\in P$ sees one guard whose colour is different from all other guards vi
sible from $z$. Such placement of colour guards is known as weak conflict-
free colouring of $P$.\n\nWe derive a lower bound of $\\Omega(\\log \\log
n)$ for weak conflict-free colouring of point guards in a simple polygon.
There is no non-trivial lower bound known for point guards. For the same p
roblem in a polygon $F$ with holes\, we also derive a lower bound of $\\Om
ega(\\log n)$\, where the guards are allowed to be placed at all locations
inside $F$ except some designated locations.\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=4224
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4224
END:VEVENT
BEGIN:VEVENT
SUMMARY:Paths to Stable Allocations
DTSTART;VALUE=DATE-TIME:20150219T083000Z
DTEND;VALUE=DATE-TIME:20150219T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4225@cern.ch
DESCRIPTION:In the stable marriage problem\, men rank women and women rank
men in a strict order of preference. The goal is to find a set of marriag
es that does not allow a blocking pair: an unmatched pair so that both the
man and the woman in this pair prefer each other to their partners. The s
table allocation problem is one of the broadest extensions of this well-kn
own problem. In an allocation problem\, edges of a bipartite graph have ca
pacities and vertices have quotas to fill. Here we investigate the case of
uncoordinated processes in stable allocation instances. In this setting\,
a feasible allocation is given and the aim is to reach a stable allocatio
n by raising the value of the allocation along edges of blocking pairs and
reducing it on worse edges if needed. Do such myopic changes lead to a st
able solution?\n\nKeywords: game theory\, graph algorithms\, matchings.\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4225
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4225
END:VEVENT
BEGIN:VEVENT
SUMMARY:Interactive Theorem Proving in Coq and the Curry-Howard Isomorphis
m
DTSTART;VALUE=DATE-TIME:20150220T053000Z
DTEND;VALUE=DATE-TIME:20150220T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4203@cern.ch
DESCRIPTION:There seems to be a general consensus among mathematicians abo
ut the notion of a correct proof. Still\, in mathematical literature\, man
y invalid proofs remain accepted over a long period of time. It happens mo
stly because proofs are incomplete\, and it is easy to make mistake while
verifying an incomplete proof. However\, writing a complete proof on paper
consumes a lot of time and effort. Proof Assistants\, such as Coq\, can m
inimize these overheads. It provides the user with lots of tools and tacti
cs to interactively develop a proof. Realization of such a computer based
tool for representing proofs and propositions\, is made possible because o
f the Curry-Howard Isomorphism. It tells us that\, proving and programming
are essentially equivalent tasks. More precisely\, we can use a programmi
ng paradigm such as typed lambda calculus to encode propositions and thei
r proofs. In this talk we will see\, how Curry-Howard Isomorphism makes it
possible to encode different logics in the Coq Proof Assistant.\n\nhttps:
//indico.tifr.res.in/indico/conferenceDisplay.py?confId=4203
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4203
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Challenges in Verification of Concurrent Programs!
DTSTART;VALUE=DATE-TIME:20150224T103000Z
DTEND;VALUE=DATE-TIME:20150224T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4235@cern.ch
DESCRIPTION:Concurrent software is everywhere. Ensuring reliability of con
current software is a daunting task. In this talk\, I will present the bas
ics of verification\, what is so hard about concurrency\, and recent effor
ts to tame the problem. The talk will also include some of my past works i
n the topic and one of my current projects.\n\nhttps://indico.tifr.res.in/
indico/conferenceDisplay.py?confId=4235
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4235
END:VEVENT
BEGIN:VEVENT
SUMMARY:Statistical Inference Based on a Parametric Family of Divergences
DTSTART;VALUE=DATE-TIME:20150302T060000Z
DTEND;VALUE=DATE-TIME:20150302T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4161@cern.ch
DESCRIPTION:We study minimization problems with respect to a one-parameter
family of generalized divergences (denoted I_{\\alpha}(P\,Q)). These I_{\
\alpha}-divergences are a generalization of the so-called Kullback-Leibler
divergence (KL-divergence). Just like KL-divergence\, these I_{\\alpha}-d
ivergences also behave like squared Euclidean distance and satisfy the Pyt
hagorean property. This talk is about the usefulness of these geometric pr
operties in robust statistics. The talk is organized in three parts.\n\nIn
the first part\, we study minimization of I_{\\alpha}(P\,Q) as the first
argument varies over a family of probability distributions that satisfy li
near statistical constraints. Such a constraint set is called a linear fam
ily. This minimization problem generalizes the maximum Renyi or Tsallis en
tropy principle of statistical physics. The structure of the minimizing pr
obability distribution naturally suggests a statistical model of power-law
probability distributions\, which we call an \\alpha-power-law family. Th
is is analogous to the exponential family that arises when relative entrop
y is minimized subject to the same linear statistical constraints.\n\nIn t
he second part\, we study minimization of I_{\\alpha}(P\,Q) over the secon
d argument. This minimization is generally on parametric families such as
the exponential family or the \\alpha-power-law family\, and is of interes
t in robust statistical estimation.\n\nIn the third part\, we show an orth
ogonality relationship between an \\alpha-power-law family and an associat
ed linear family. As a consequence of this\, the minimization of I_{\\alph
a} over the second argument on an \\alpha-power-law family can be shown to
be equivalent to a minimization of I_{\\alpha} over the first argument on
a linear family. The latter turns out to be a simpler problem of minimiza
tion of a quasi-convex objective function subject to linear constraints (t
his is a joint work with Rajesh Sundaresan).\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=4161
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4161
END:VEVENT
BEGIN:VEVENT
SUMMARY:Greedy Recovery Algorithms in Compressive Sensing: A Review and So
me New Results
DTSTART;VALUE=DATE-TIME:20150302T090000Z
DTEND;VALUE=DATE-TIME:20150302T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4248@cern.ch
DESCRIPTION:Compressed sensing or compressive sampling (CS) is a powerful
technique to represent signals at a sub-Nyquist sampling rate while retain
ing the capacity of perfect (or near perfect) reconstruction of the signal
\, provided the signal is known to be sparse in some domain. In last few y
ears\, the CS technique has attracted considerable attention from across a
wide array of fields like applied mathematics\, statistics\, and engineer
ing\, including signal processing areas like MR imaging\, speech processin
g\, analog to digital conversion etc. The framework of CS essentially lead
s to finding the sparsest solution to a set of under-determined linear equ
ations\, say\, y = A x \, where A is a M by N sensing matrix and y is a c
ompressed measurement vector. The ideal approach to find the sparsest solu
tion is based on minimization of the l0 norm of x under the condition y =
A x . However\, due to non-convexity of the l0 norm\, this leads to a N
P hard problem and is thus not practical. It has\, however\, been shown th
at one can obtain the same sparsest solution by replacing the l0 norm of
x by its l1 norm provided the matrix A satisfies a so-called restricted
isometry property (RIP). In recent years\, a class of algorithms called gr
eedy CS recovery algorithms have come up that exploit the RIP and evaluate
the above stated sparsest solution by iteratively constructing its true s
upport.\n\nThis talk will introduce the basics of compressed sensing to th
e audience and take a review of some of the well known greedy recovery alg
orithms. This will then be followed up by a presentation of some new resul
ts related to convergence of some of these algorithms.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=4248
LOCATION: AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4248
END:VEVENT
BEGIN:VEVENT
SUMMARY:Blackwell's Approachability and Online Convex Optimization: Some N
ew Relations
DTSTART;VALUE=DATE-TIME:20150303T103000Z
DTEND;VALUE=DATE-TIME:20150303T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4229@cern.ch
DESCRIPTION:David Blackwell introduced in 1956 the problem of set-approach
ability in repeated games with vector payoffs\, along with geometric appro
achability conditions and corresponding strategies that rely on computing
"steering directions" as projections from the current average-payoff vecto
r to the target set. An alternative framework for generating these steerin
g directions has recently been proposed\, which relies on the modern theor
y of no-regret algorithms and online convex optimization\, along with some
basic relations from convex analysis. In this talk we will describe this
general framework\, derive some concrete algorithms along with their perfo
rmance\, and finally explore some useful relations to Blackwell's original
algorithm. The required background on approachability and on Online Conve
x Optimization will be provided as well.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=4229
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4229
END:VEVENT
BEGIN:VEVENT
SUMMARY:Automatic Speech Recognition
DTSTART;VALUE=DATE-TIME:20150305T083000Z
DTEND;VALUE=DATE-TIME:20150305T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4255@cern.ch
DESCRIPTION:ASR or Automatic Speech Recognition is the process of conversi
on of spoken utterance by humans into text by machines. Currently used ASR
tools use probabilistic techniques such as HMM (Hidden Markov Models) for
modeling sounds and then employ pattern matching to decode an unknown spe
ech utterance. The talk will highlight the major difficulties in the ASR p
rocess and how different algorithms\, techniques try to overcome these dif
ficulties.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=4255
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4255
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hardness of Approximate Coloring
DTSTART;VALUE=DATE-TIME:20150310T103000Z
DTEND;VALUE=DATE-TIME:20150310T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4239@cern.ch
DESCRIPTION:The graph coloring problem is a notoriously hard problem for w
hich we do not have efficient algorithms. A $k$-coloring of a graph is an
assignment of "colors" from $\\{1\,\\cdots\, k\\}$ such that the end point
s of every edge have different colors. Given a 3-colorable graph\, the bes
t known efficient algorithms output a $n^{0.2}$-coloring. It is known that
efficient algorithms cannot find a $4$-coloring. Hence there is a large
gap ($n^{0.2}$ vs 4) between what current algorithms can achieve and the
hardness results known.\n \nIn this thesis\, we improve the hardness resul
ts for graph coloring and its generalizations. We show the following resul
ts:\n \n1. For the case of (almost) 3-colorable graph\, we show hardness o
f finding a $2^{poly(log log n)}$-coloring\, assuming a variant of the Uni
que Games Conjecture (UGC). \n \n2. For the case of 4-colorable 4-uniform
hypergraphs\, we show hardness of finding a $2^{(\\log n)^{1/18}}$-colorin
g. \n \n3. For the problem of the approximating the covering number of CSP
s with non-odd predicates\, we show hardness of approximation to any cons
tant factor\, assuming a variant of UGC.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=4239
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4239
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Martin's Determinacy of Borel Games and its Application to Stoc
hastic Games
DTSTART;VALUE=DATE-TIME:20150311T060000Z
DTEND;VALUE=DATE-TIME:20150311T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4222@cern.ch
DESCRIPTION:While every win-lose two person extensive game with finitely m
any moves and finitely many actions in each move and with perfect informat
ion admits an optimal winning strategy\, it can fail to be true once the n
umber of moves is countable even if the action sets for the players are fi
nite in each move. Gale and Stewart introduced this class of games and sho
wed that open sets or closed sets defined on the terminal vertices of the
infinite game tree as winning sets admit winning strategies and hence thos
e games are determined. Blackwell showed that $G_{\\delta}$ set as winnin
g set on the terminal vertices are also determined. Martin proved the rema
rkable theorem that if the winning set is a Borel subset of the terminal v
ertices\, then also such games are determined.\n\nWhile zero sum two perso
n stochastic games with finitely many states and actions admit stationary
optimal strategies for discounted payoffs\, the existence of value with Ce
saro payoff is possible only in the space of behavioral strategies. At eac
h move players may have to peg on the entire history so far to make their
randomized action choices. The seminal theorem of Mertens and Neyman on th
e existence of value for Cesaro payoff stochastic games is made quite simp
le and transparent by an application of the above theorem of Martin on per
fect information Gale Stewart games. It simply bypasses many complicated c
onstructions and technical estimates that view solution to the discounted
Shapley value equation as an elementary sentence in the space of the order
ed field of Laurentz series in fractional powers of the discount factor\,
namely the real closed field of Puiseux series and hence relies on many to
ols from real algebraic geometry. The talk will present this proof due to
Ashok Maitra and Sudderth. Their theorem is applicable to much more gener
al classes of payoffs besides Cesaro payoffs. This is a fruitful interplay
between mathematical logic and game theory.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=4222
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4222
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Algorithm for Finding the Nucleolus of Assignment Games
DTSTART;VALUE=DATE-TIME:20150312T103000Z
DTEND;VALUE=DATE-TIME:20150312T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4223@cern.ch
DESCRIPTION:Assignment games with side payments are models of certain two-
sided markets. It is known that prices which competitively balance supply
and demand correspond to elements in the core. The nucleolus\, lying in th
e lexicographic center of the nonempty core\, has the additional property
that it satisfies each coalition as much as possible. The corresponding pr
ices favor neither the sellers nor the buyers\, hence provide some stabili
ty for the market.\n\nAn algorithm is presented that determines the nucleo
lus of an assignment game. It generates a finite numberof payoff vectors\,
monotone increasing on one side\, and decreasing on the other. The decomp
osition of the payoff space and the lattice-type structure of the feasible
set are utilized in associating a directed graph. Finding the next payoff
is translated into determining the lengths of longest paths to the nodes\
, if the graph is acyclic\, or otherwise\, detecting the cycle(s). In an $
(m\, n)$-person assignment game with $m = min(m\, n)$ the nucleolus is fou
nd in at most 1/2 $m(m+3)$ steps\, each one requiring at most $O(m n)$ ele
mentary operations.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=4223
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4223
END:VEVENT
BEGIN:VEVENT
SUMMARY:Topology in Combinatorics: The Borsuk-Ulam Theorem
DTSTART;VALUE=DATE-TIME:20150312T083000Z
DTEND;VALUE=DATE-TIME:20150312T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4269@cern.ch
DESCRIPTION:We will see an application of topology to prove combinatorial
results. The Borsuk-Ulam theorem states that any continuous function f fro
m the surface of the (n+1)-sphere to R^n should have a point x on the sphe
re such that f(x) = f(-x). For instance when n=2\, it implies that if one
deflates a football and places it flattened out (arbitrarily) on a table\,
then there must be two *diametrically opposite* points that land up on to
p of each other on the table.\n\nPerhaps surprisingly\, this theorem can b
e used to prove the fact that the Kneser-Graph (n\,k) has chromatic number
exactly n-2k+2 (conjectured by Kneser\, proven first by Lovasz in 1978).
We will see this and a couple more applications of (variants of) the Borsu
k-Ulam theorem in the talk. For some of these results\, no purely combinat
orial proofs are known.\n\nhttps://indico.tifr.res.in/indico/conferenceDis
play.py?confId=4269
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4269
END:VEVENT
BEGIN:VEVENT
SUMMARY:Old and New PCP Constructions
DTSTART;VALUE=DATE-TIME:20150317T103000Z
DTEND;VALUE=DATE-TIME:20150317T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4273@cern.ch
DESCRIPTION:The PCP theorem (AS\,ALMSS 1991) guarantees that every NP lang
uage has a probabilistically checkable proof (PCP) system allowing a verif
ier to check a witness very efficiently using randomness\, and allowing fo
r small error.\n\nMost of the talk will not assume prior knowledge\, but I
will also devote some time to some recent work joint with Dinur and Kindl
er.\n\nIn this work we make (some) progress towards proving the so-called
"sliding-scale conjecture". This is a conjecture of Bellare-Goldreich-Lund
-Russel from 1993 about the tradeoff between the number of bits read from
the PCP proof and the error of the verifier. Our work revisits older const
ructions and analyzes them using the more modern "modular-composition" app
roach (based on joint work with Irit Dinur and Guy Kindler).\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=4273
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4273
END:VEVENT
BEGIN:VEVENT
SUMMARY:Constant Depth Circuit Complexity of Powering
DTSTART;VALUE=DATE-TIME:20150318T104500Z
DTEND;VALUE=DATE-TIME:20150318T114500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4272@cern.ch
DESCRIPTION:Boolean circuit classes inherently capture the power of parall
el computation\, and constant depth circuits are a class of great interest
. It is a classical result that the parity of $n$ bits cannot be computed
by a constant depth polynomial sized Boolean circuit with unbounded fan-in
AND and OR gates ($AC^0$)\n\nA natural question is how the power of these
circuits increase if we add parity gates to them. Arithmetic over $\\mat
hbb{F}_{2^n}$ is naturally captured by this class since multiplication can
be computed by small circuits\, and addition by parity gates. One of the
most basic arithmetic operations is powering.\n\nIn this report\, we explo
re the complexity of computing the $k$th power of an element $x \\in \\mat
hbb{F}_{2^n}$ by $AC^0$ circuits equipped with parity gates. In particula
r\, we try to prove hardness in two cases\; one where the binary represent
ation of the exponent looks like a periodic string\, and the other where t
he exponent has large number of 0-1 alternations in its binary representat
ion.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4272
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4272
END:VEVENT
BEGIN:VEVENT
SUMMARY:Equilibrium in Markets: Algorithms and Complexity
DTSTART;VALUE=DATE-TIME:20150319T053000Z
DTEND;VALUE=DATE-TIME:20150319T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4258@cern.ch
DESCRIPTION:Market equilibrium is a central solution concept in economics\
, where prices are set such that the demand and supply are balanced. It ha
s been under study ever since the inception of the field starting with the
pioneering work of Adam Smith (1776) and Leon Walras (1874). In a celebra
ted result\, Arrow and Debreu (1954) showed the existence of equilibrium u
sing the Kakutani fixed point theorem\, which is highly non-constructive.
Since then many algorithms have been proposed\, especially worth mentionin
g are algorithms of Scarf (1967) and Smale (1976) to find approximate fixe
d points. While much progress has been made in the last century\, computat
ionally large part of progress happened in last fifteen years using tools
from complexity theory\, optimization and mathematical programming. This h
as led to a remarkable theory of computability of equilibria in markets. M
y work touches many aspects of this problem from computation to strategic
analysis to learning. While the problem has turned out to be intractable f
or most of the important cases\, designing even a non-enumerative algorith
m for these cases remained elusive. I will show the first "practical" algo
rithms for a very general class of markets settling long standing open que
stions. These algorithms are based on the powerful machinery of linear com
plementarity problem (LCP) and complementary pivot (Simplex-like) algorith
ms. I will also show how these results help bring out a clear dichotomy am
ong equilibrium computation problems.\n\nBio: Jugal Garg is currently a re
search fellow at Max-Planck-Institute for Informatics\, Germany. Prior to
that\, he was a Algorithms and Randomness Center (ARC) postdoctoral fellow
at Georgia Tech for two years. He received his PhD from IIT-Bombay in 201
2. Jugal's research explores computational and strategic aspects of equili
bria in game theory and economics\, and their connections with dynamical s
ystems and learning. He is interested broadly in design and analysis of al
gorithms\, optimization and mathematical programming.\n\nhttps://indico.ti
fr.res.in/indico/conferenceDisplay.py?confId=4258
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4258
END:VEVENT
BEGIN:VEVENT
SUMMARY:Pure State Transformations Under Local Operations and Classical Co
mmunication (LOCC)
DTSTART;VALUE=DATE-TIME:20150326T083000Z
DTEND;VALUE=DATE-TIME:20150326T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4288@cern.ch
DESCRIPTION:We discuss Nielsen's result(1999) which characterizes pure sta
te transformation under LOCC by connecting the quantum entanglement to the
algebraic theory of majorization.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=4288
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4288
END:VEVENT
BEGIN:VEVENT
SUMMARY:Studying Triangulations and Visibility Graphs of Planar Point Sets
DTSTART;VALUE=DATE-TIME:20150407T090000Z
DTEND;VALUE=DATE-TIME:20150407T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4291@cern.ch
DESCRIPTION:Point sets on the plane are a fundamental class of objects stu
died in discrete and computational geometry\, with many applications in re
al life problems. In this seminar\, we discuss the following three problem
s on planar point sets.\n \nA triangulation of a planar point set P is a p
lane graph in which every face (except the outer face) is a triangle. A pl
anar point set P is k-connectible if there exists a k-connected plane grap
h G with vertex set P and all edges as line segments. We characterize all
4-connectible point sets and give an O(n^3) time algorithm to compute such
a triangulation. Thus\, we solve a longstanding open problem in computati
onal geometry and geometric graph theory.\n \nTwo points p and q of a poin
t set P are mutually visible if the line segment pq does not contain or pa
ss through any other point of P. The point visibility graph G of P (referr
ed to as PVG) is defined by associating a vertex with each point of P such
that (v\,w) is an undirected edge of G iff their corresponding points p a
nd q are visible. Identifying a set of properties satisfied by all PVGs is
called the PVG characterization problem. We give three necessary conditio
ns for characterizing PVGs\, and discuss properties of PVGs. We completely
characterize planar PVGs.\n \nThe problem of determining if there is a se
t of points P whose PVG is the given graph G\, is called the visibility gr
aph recognition problem. We show that the PVG recognition problem is NP-ha
rd using collinearity restrictions for points.\n\nhttps://indico.tifr.res.
in/indico/conferenceDisplay.py?confId=4291
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4291
END:VEVENT
BEGIN:VEVENT
SUMMARY:Broadcast Channels and Their Capacity Under Noisy Feedback
DTSTART;VALUE=DATE-TIME:20150420T103000Z
DTEND;VALUE=DATE-TIME:20150420T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4327@cern.ch
DESCRIPTION:Broadcast channels model the downlink in a wireless network\,
i.e.\, the link from a base station transmitter to several receivers. The
broadcast channel model was proposed by Tom Cover in 1972. While character
izing the capacity region of the broadcast channel remains an important op
en problem in Information Theory\, the capacity of the Gaussian version of
this channel has been resolved (single antenna version in 1974 and multi-
antenna version in 2005).\n\nFeeding signals received from the channel by
the receiver(s) back to the transmitter(s) may\, in general\, lead to an e
nlargement of the capacity region of multi-terminal channels. A famous exc
eption to this rule is the memoryless point-to-point channel (i.e.\, a sin
gle transmitter & a single receiver). For a limited class of memoryless br
oadcast channels\, called the physically degraded broadcast channels\, fee
dback does not enlarge the capacity region (El Gamal\, '78\,'81). However\
, this class does not contain wireless channels which arise in practice.\n
\nResults available in the literature show that feedback of even a limited
form can result in an enlargement of the capacity region of memoryless br
oadcast channels. Full feedback (Dueck '80\, Ozarow-Leung '84)\, feedback
from only one receiver (Pillai '08)\, noisy feedback (Shayevitz-Wigger '13
\, Venkataramanan-Pradhan '13)\, rate-limited feedback (Yu-Wigger '14) hav
e all been shown to lead to an enlargement of the capacity region in chann
els of practical interest. For multiple access channels (the uplink analog
ue of broadcast channels)\, it is known that even partial noisy feedback a
lways enlarges the capacity region. A duality which is known to exist betw
een multiple access and broadcast channels has led to the speculation that
a similar result may hold for broadcast channels. We show that this is no
t the case in a class of channels which includes channels which arise in p
ractice. In particular\, we identify a threshold for the noise power in th
e feedback link above which feedback from a single receiver does not enlar
ge the capacity region of the memoryless Gaussian broadcast channel with i
ndependent noises (joint work with Sibiraj B Pillai (IIT Bombay).\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4327
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4327
END:VEVENT
BEGIN:VEVENT
SUMMARY:Networks in Corporate Finance: Some Applications and Research Oppo
rtunities
DTSTART;VALUE=DATE-TIME:20150425T053000Z
DTEND;VALUE=DATE-TIME:20150425T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4326@cern.ch
DESCRIPTION:Networks such as LinkedIn and Facebook have spurred enormous a
nd interdisciplinary research in connections and networks. The talk focuse
s on selected applications of connectedness and network methods in corpora
te finance research. Specific topics include using text data to recalibrat
e competition among firms\, using spatial methods to characterize competit
ion between mutual funds\, using community detection methods to understand
partnerships between venture capitalists\, calibrating financial institut
ion interconnectedness to understand systemic risk\, and some insights on
whether the type of connection rather than the connection matters using pe
er to peer lending data and data on corporate frauds.\n\nhttps://indico.ti
fr.res.in/indico/conferenceDisplay.py?confId=4326
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4326
END:VEVENT
BEGIN:VEVENT
SUMMARY:Improved Sample Complexity Estimates for Stochastic Approximation
Algorithms
DTSTART;VALUE=DATE-TIME:20150430T083000Z
DTEND;VALUE=DATE-TIME:20150430T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4343@cern.ch
DESCRIPTION:Abstract: The Alexeev's perturbation theory gives a method to
compare solutions of a perturbed ordinary differential equation (ODEs) wit
h that of the unperturbed one. In this talk\, we shall discuss an approach
based on this method to obtain sample complexity estimates of stochastic
approximation (SA) algorithms. Note that sample complexity refers to the p
robability that the SA iterates $x_{n}$ are within an $\\epsilon-$neighbou
rhood of the equilibrium after lapse of a certain amount of time\, conditi
oned on the event that $x_{n_0}$ was in some bigger neighbourhood of the e
quilibrium (this is a working paper with Prof. V. Borkar).\n\nhttps://indi
co.tifr.res.in/indico/conferenceDisplay.py?confId=4343
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4343
END:VEVENT
BEGIN:VEVENT
SUMMARY:Introduction and Some Properties of Unitary t-designs
DTSTART;VALUE=DATE-TIME:20150507T083000Z
DTEND;VALUE=DATE-TIME:20150507T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4355@cern.ch
DESCRIPTION:When studying ''random" operators it is essential to be able t
o integrate over the Haar measure. Unitary t-designs which have many appli
cations in Quantum information theory provide a method to simplify integra
ting polynomials of degree less than 't' over U(d) by replacing the averag
es over haar measure by the averages over a finite set. We see an explicit
construction of a 1-design and use it to prove that t-designs are non-com
muting.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4
355
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4355
END:VEVENT
BEGIN:VEVENT
SUMMARY:Models and Algorithms for Database Segment Location in Information
Networks
DTSTART;VALUE=DATE-TIME:20150511T103000Z
DTEND;VALUE=DATE-TIME:20150511T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4358@cern.ch
DESCRIPTION:In a content delivery network\, data are distributed to geogra
phically separated servers or data centers. This helps in distributing the
load in the network and in serving the neighborhood locally. However\, an
arbitrary data placement scheme may lead to excessive operational cost fo
r the service provider. We study the problem of content distribution and d
elivery as a cost optimization problem.\n\nThe major decisions influencing
the cost involve the location of servers\, allocation of content to serve
rs\, assignment of users to servers\, the network topology and routing cho
ices. Clearly\, these decisions also appear in a transportation network de
sign. We use the concepts of hub-spoke networks\, widely used in the airli
nes and telecommunications industry\, to model these problems in an integr
ated framework. We introduce a two-phased solution approach\, in which the
first phase identifies a set of disjoint parts of the databases (called d
atabase segments) by using ideas from collaborative filtering. In the seco
nd phase\, the optimal placement of these segments in a network is studied
in a mathematical programming framework. Due to the presence of multiple
problems\, some of which are already known to be NP-hard\, the resulting f
ormulation is extremely complex and the optimal solutions are available on
ly for small instances. Therefore\, we focus on the development of large s
cale optimization techniques (such as Benders decomposition) to make the m
odel applicable in practice.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=4358
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4358
END:VEVENT
BEGIN:VEVENT
SUMMARY:Broadcast Channels and Their Capacity Under Noisy Feedback
DTSTART;VALUE=DATE-TIME:20150514T103000Z
DTEND;VALUE=DATE-TIME:20150514T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4359@cern.ch
DESCRIPTION:Broadcast channels model the downlink in a wireless network\,
i.e.\, the link from a base station transmitter to several receivers. The
broadcast channel model was proposed by Tom Cover in 1972. While character
izing the capacity region of the broadcast channel remains an important op
en problem in Information Theory\, the capacity of the Gaussian version of
this channel has been resolved (single antenna version in 1974 and multi-
antenna version in 2005).\n\nFeeding signals received from the channel by
the receiver(s) back to the transmitter(s) may\, in general\, lead to an e
nlargement of the capacity region of multi-terminal channels. A famous exc
eption to this rule is the point-to-point channel (i.e.\, a single transmi
tter & a single receiver). For a limited class of broadcast channels\, cal
led the physically degraded broadcast channels\, feedback does not enlarge
the capacity region (El Gamal\, '78\,'81). However\, this class does not
contain wireless channels which arise in practice.\n\nResults available in
the literature show that feedback of even a limited form can result in an
enlargement of the capacity region of broadcast channels. Full feedback (
Dueck '80\, Ozarow-Leung '84)\, feedback from only one receiver (Pillai '0
8)\, noisy feedback (Shayevitz-Wigger '13\, Venkataramanan-Pradhan '13)\,
rate-limited feedback (Yu-Wigger '14) have all been shown to lead to an en
largement of the capacity region in channels of practical interest. For mu
ltiple access channels (the uplink analogue of broadcast channels)\, it is
known that even partial noisy feedback always enlarges the capacity regio
n. A duality which is known to exist between multiple access and broadcast
channels has led to the speculation that a similar result may hold for br
oadcast channels. We show that this is not the case in a class of channels
which includes channels which arise in practice. In particular\, we ident
ify a threshold for the noise power in the feedback link above which feedb
ack from a single receiver does not enlarge the capacity region (joint wor
k with Sibiraj B Pillai (IIT Bombay).\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=4359
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4359
END:VEVENT
BEGIN:VEVENT
SUMMARY:Clique vs Independent Set Game: A Tight Bound
DTSTART;VALUE=DATE-TIME:20150515T083000Z
DTEND;VALUE=DATE-TIME:20150515T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4364@cern.ch
DESCRIPTION:The clique vs independent set game (CIS(G)) on a graph $G$ is
a game between two parties - Alice and Bob. Alice get s a set of vertices\
, $C$\, which forms a clique in $G$ and Bob gets a set of vertices\, $I$\
, which forms an independent set in $G$. Clearly\, $C$ and $I$ can interse
ct in only 1 vertex. The question the players try to answer is whether $C
\\cap I = \\emptyset$ or not.\n\nBy a result of Yannakakis\, we know that
this problem can be solved deterministically by communicating $O(\\log^2 n
)$ bits where $n$ is the size of vertex set in $V$. It has been an open pr
oblem for a long time whether this upper bound can be improved.\n\nA recen
t result of Goos\, Pitassi and Watson shows that this upper bound is tight
\, i.e.\, there is a graph for which the CIS game requires $\\tilde{\\Omeg
a}(\\log^2 n)$ communication. In the talk\, we will see proof-sketches of
these results.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=4364
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4364
END:VEVENT
BEGIN:VEVENT
SUMMARY:Art Gallery Problems for Visibility with Reflection
DTSTART;VALUE=DATE-TIME:20150529T083000Z
DTEND;VALUE=DATE-TIME:20150529T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4383@cern.ch
DESCRIPTION:The art gallery problem is to determine the minimum number of
point light sources positioned at vertices of a polygon that can illuminat
e the entire polygon. Lee and Lin showed that this problem is NP-hard for
direct visibility. In this talk\, we will show that this problem remains N
P-hard even for visibility with reflection\, considering both *specular* a
nd *diffuse* reflections. If time permits\, I'll also talk about a variant
of the art gallery problem known as the kernel problem.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=4383
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4383
END:VEVENT
BEGIN:VEVENT
SUMMARY:Two Player Games and Playing Them in Parallel
DTSTART;VALUE=DATE-TIME:20150529T103000Z
DTEND;VALUE=DATE-TIME:20150529T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4384@cern.ch
DESCRIPTION:Consider a 2-player game defined as follows. A referee sends a
pair of questions x and y to two players Alice and Bob and accepts if the
ir answers a and b satisfy some predicate. The 2 players are not allowed t
o communicate once the game commences\, however can share some prior rando
mness\, quantum entanglement etc. We will begin by seeing examples of game
s in which the players have different optimal strategies depending on what
they share: randomness\, entanglement\, no-signalling strategies\, etc.\n
\nWe will then consider the parallel repetition of games and ask if the pl
ayers have an advantage over answering each of the repetitions independent
ly. Surprisingly\, we will show examples of games in which the players hav
e an advantage by co-ordinating their answers across repetitions. Finally\
, we will show time permitting\, we will see some applications of parallel
repetition to theoretical computer science.\n\nImage from: Andreas Winter
\, "Quantum mechanics: The usefulness of uselessness"\, Nature 466\, 1053
–1054\, Aug 2010.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=4384
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4384
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Stochastic Approximation\, Stochastic Algebraic Topology\, and
Optimization in High Dimensions
DTSTART;VALUE=DATE-TIME:20150611T053000Z
DTEND;VALUE=DATE-TIME:20150611T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4389@cern.ch
DESCRIPTION:In this talk\, we shall discuss three problems respectively fr
om stochastic approximation\, stochastic algebraic topology\, and optimiza
tion in high dimensions.\n\n\\textbf{Problem 1:} Stochastic approximation
(SA) refers to algorithms that attempt to find optimal points or zeroes of
a function when only its noisy estimates are available. Given $\\epsilon
> 0\,$ sample complexity of SA algorithm is the probability that its itera
tes are eventually within an $\\epsilon-$neighbourhood of a desired soluti
on\, conditioned on the event that the iterates did enter some bigger neig
hbourhood of this desired solution. In the first part of this talk\, we sh
all obtain a tight sample complexity estimate by making use of the Aleksee
v's analogue of variation of constants formula for nonlinear systems and a
generalization of a concentration result from Liu and Watbled\, 2009.\n\n
\\textbf{Problem 2:} In stochastic algebraic topology\, one is interested
in the algebraic topology of random sets. In the second part of this talk\
, we shall discuss a time varying analogue of the Erdos-Renyi graph\, whic
h we call the dynamic Erdos-Renyi graph\, and concentrate on the topologic
al aspects of its clique complex. Each edge of the dynamic Erdos-Renyi gra
ph independently evolves as a continuous time on/off Markov chain such tha
t the connection probability is $p$ at all times. Our main result is the w
eak convergence to stationary Ornstein-Uhlenbeck process of a (normalized)
random process determined by $k-$th Betti number of the clique complex as
sociated with the graph when $p=n^\\alpha\,$ with $\\alpha \\in (−1/k\,
−1/(k+1)).$\n\n\\textbf{Problem 3:} High dimensional unconstrained quadr
atic programs (UQPs) involving massive datasets are now common. Without co
mputational resources that match up to these datasets\, solving such probl
ems using classical UQP methods is very difficult. In the third part of th
is talk\, we shall first define high dimensional compliant (HDC) methods f
or UQPs---methods that can solve high dimensional UQPs by adapting to avai
lable computational resources. We will then show that the class of block K
aczmarz and block coordinate descent (BCD) are the only existing methods t
hat can be easily made HDC. We shall then discuss a novel deterministic\,
but adaptive\, greedy BCD (GBCD) method for high dimensional UQPs. We shal
l also see theoretical and experimental tests that demonstrate the effecti
veness of GBCD.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=4389
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4389
END:VEVENT
BEGIN:VEVENT
SUMMARY:Polygon Guarding With Orientation
DTSTART;VALUE=DATE-TIME:20150612T083000Z
DTEND;VALUE=DATE-TIME:20150612T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4402@cern.ch
DESCRIPTION:The art gallery problem is a classical sensor placement proble
m that asks for the minimum number of guards required to see every point i
n an environment. The standard formulation does not take into account self
-occlusions caused by a person or an object within the environment. Obtain
ing good views of an object from all orientations despite self-occlusions
is an important requirement for surveillance and visual inspection applica
tions. In this talk\, we will see a version of the art gallery problem und
er a constraint\, termed △-guarding\, that ensures that all sides of any
convex object are always visible in spite of self-occlusion.\n\nI will pr
esent 3 main results in this talk. First\, I will prove that Omega(√n) g
uards are always necessary for △-guarding the interior of a simple polyg
on having n vertices. Second\, I will present a O(log c_{opt}) factor appr
oximation algorithm for △-guarding polygons with or without holes\, when
the guards are restricted to vertices of the polygon\, where c_{opt} is t
he optimal number of guards. Third\, I will introduce the problem of △-g
uarding a set of line segments connecting points on the boundary of the po
lygon. This is motivated by applications where an object or person of inte
rest can only move along certain paths in the polygon. If time permits\, I
will outline a constant factor approximation algorithm for this problem
– one of the few such results for art gallery problems.\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=4402
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4402
END:VEVENT
BEGIN:VEVENT
SUMMARY:Axiomatizing Equality of Expressions
DTSTART;VALUE=DATE-TIME:20150618T103000Z
DTEND;VALUE=DATE-TIME:20150618T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4407@cern.ch
DESCRIPTION:An axiomatization of equality of regular expressions was given
by Arto Salomaa in 1966\, using Janusz Brzozowski's "derivatives" in 1964
. Volodimir Redko also showed in 1964 that a finite set of axioms is insuf
ficient\, so Salomaa used an algorithmic side condition in his system. Joh
n Horton Conway formulated an algebraic approach in 1971. Albert Meyer and
Larry Stockmeyer showed in 1972 that the problem requires polynomial spac
e. Dexter Kozen developed a new "inequational" axiomatization using Conway
's algebraic ideas in 1994 (requiring no algorithmic side conditions). Jan
Rutten suggested using "coalgebraic" ideas in 1999 and Clemens Grabmayer
came up with an axiomatization in 2005. This talk surveys these ideas and
points out challenges remaining after 50 years of work in this area.\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4407
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4407
END:VEVENT
BEGIN:VEVENT
SUMMARY:Monte Carlo Methods in Finance
DTSTART;VALUE=DATE-TIME:20150618T103000Z
DTEND;VALUE=DATE-TIME:20150618T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4411@cern.ch
DESCRIPTION:In the early 70’s\, Black and Scholes introduced the now fam
ous no-arbitrage based techniques for pricing exotic financial securities.
This led financial mathematics to move away from business and economics d
epartments to mathematics and engineering. Around the same time\, Monte Ca
rlo methods (roughly – throwing darts to approximately measure areas of
regions\, but through fancy computer algorithms) were expanding their util
ity beyond physics and other sciences to engineering and business applicat
ions. The two met – some say that the resulting marriage was preordained
in the no-arbitrage principle.\n\nSo this talk is about love\, marriage\,
destiny and lots of money\, with a little bit of tragedy in the form of t
he last financial crisis.\n\nIf time permits I will also discuss applicati
ons of the appropriately termed multi-armed bandit techniques\, popular in
online learning\, to Monte Carlo based financial portfolio risk measureme
nt.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4411
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4411
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stochastic Averaging for Randomly Perturbed Hamiltonian Systems
DTSTART;VALUE=DATE-TIME:20150623T103000Z
DTEND;VALUE=DATE-TIME:20150623T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4403@cern.ch
DESCRIPTION:The technique of averaging (deterministic or stochastic) is us
ed to effect model reduction in systems which possess dynamics on multiple
time scales. Roughly speaking\, the idea is to obtain reduced models for
slowly varying quantities by taking long-term averages in rapidly varying
quantities. In this talk\, we overview some problems in stochastic averagi
ng for a class of planar Hamiltonian systems subjected to small white nois
e perturbations. Under a suitable time-rescaling\, the dynamics can be see
n to comprise a fast rotation along orbits of the Hamiltonian flow togethe
r with a slow transversal diffusion across orbits. As the ratio of speeds
of the slow to fast variables goes to zero\, the limiting motion is given
by a graph-valued Markov process governed by linear second-order differen
tial operators on the legs of the graph together with so-called glueing co
nditions at each vertex. We present some recent results for the case of a
planar Hamiltonian system with skew random perturbations.\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=4403
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4403
END:VEVENT
BEGIN:VEVENT
SUMMARY:Managing Congestion in Dynamic Matching Markets
DTSTART;VALUE=DATE-TIME:20150720T103000Z
DTEND;VALUE=DATE-TIME:20150720T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4452@cern.ch
DESCRIPTION:We consider a decentralized two-sided matching market in which
agents arrive and depart asynchronously. As a result\, it is possible tha
t an agent on one side of the market (a “buyer”) identifies an agent o
n the other side of the market (a “seller”) who is a suitable match\,
only to find that the seller is already matched. We find using a mean fiel
d approach that lack of knowledge about availability can create large welf
are losses to both buyers and sellers. We consider a simple intervention a
vailable to the platform: limiting visibility of sellers. We find that thi
s intervention can significantly improve the welfare of agents on both sid
es of the market\; sellers pay lower application costs\, while buyers are
less likely to find that the sellers they screen have already matched. Som
ewhat counterintuitively\, the benefits of showing fewer sellers to each b
uyer are greatest in markets in which there is a shortage of sellers (join
t work with Nick Arnosti and Ramesh Johari).\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=4452
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4452
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fully Dynamic $(1+\\epsilon)$-Approximate Matchings
DTSTART;VALUE=DATE-TIME:20150723T103000Z
DTEND;VALUE=DATE-TIME:20150723T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4454@cern.ch
DESCRIPTION:We study data structures that maintain approximate maximum mat
chings in graphs under edge insertions/deletions. Our main result is a dat
a structure that maintains a matching whose size is at least $(1 - \\epsil
on)$ of the maximum in worst case $O(\\sqrt{m}\\epsilon^{-2})$ per update.
It is the first data structure that is able to maintain arbitrary quality
approximations on sparse graphs in sublinear time per update.\n\nOur resu
lts of maximum cardinality matching easily extend to maximum weighted matc
hing. Using known schemes\, we first obtain a $(3+\\epsilon)$ approximatio
n of maximum weighted matching with $O( \\sqrt m \\epsilon^{-2} \\log N)$
update time. Using intricate rounding schemes\, we then obtain a $(1+\\eps
ilon)$ approximation of maximum matching in $O( \\sqrt{m} \\epsilon^{-2 -
O(\\epsilon^{-1})} \\log N)$ update time. It is the first data-structure w
hich maintains arbitrary quality approximation on a weighted graph.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4454
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4454
END:VEVENT
BEGIN:VEVENT
SUMMARY:Community Detection in Networks: SDP Relaxations and Computational
Gaps
DTSTART;VALUE=DATE-TIME:20150728T090000Z
DTEND;VALUE=DATE-TIME:20150728T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4495@cern.ch
DESCRIPTION:This talk focuses on the problem of finding the underlying com
munities within a network using only knowledge of network topology. We con
sider a generative model for a network\, namely the planted cluster model\
, which is a simple extension of the classical stochastic block model. We
derive a semidefinite programming (SDP) relaxation of the maximum likeliho
od estimator for recovering the planted clusters from the network. If the
size of the community is linear in the total number of vertices\, the perf
ormance guarantee of the SDP exactly matches the necessary information bou
nd. However\, if the community size is sub-linear in the total number of v
ertices\, the performance guarantee of the SDP is far from the information
limit. Building on average case reductions\, we show there exists a signi
ficant gap between the information limit and what can be achieved by compu
tationally efficient procedures\, conditioned on the assumptions that cert
ain instances of the planted clique problem cannot be solved in randomized
polynomial time (based on joint work\, available at 1406.6625\, 1412.6156
and 1502.07738\, with Bruce Hajek (UIUC) and Jiaming Xu (Wharton)).\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4495
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4495
END:VEVENT
BEGIN:VEVENT
SUMMARY:Ordinal Optimization - Empirical Large Deviations Rate Estimators\
, and Stochastic Multi-armed Bandits
DTSTART;VALUE=DATE-TIME:20150810T090000Z
DTEND;VALUE=DATE-TIME:20150810T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4515@cern.ch
DESCRIPTION:Abstract: Consider the ordinal optimization problem of finding
a population amongst many with the smallest mean when these means are unk
nown but population samples can be generated via simulation. Typically\, b
y selecting a population with the smallest sample mean\, it can be shown t
hat the false selection probability decays at an exponential rate. Lately
researchers have sought algorithms that guarantee that this probability is
restricted to a small $\\delta$ in order $\\log(1/\\delta)$ computation
al time by estimating the associated large deviations rate function via si
mulation. We show that such guarantees are misleading. Enroute\, we identi
fy the large deviations principle followed by the empirically estimated la
rge deviations rate function that may also be of independent interest. Fur
ther\, we show a negative result that when populations have unbounded supp
ort\, any policy that asymptotically identifies the correct population wi
th probability at least $1-\\delta$ for each problem instance requires mor
e than $O(\\log(1/\\delta))$ samples in making such a determination in an
y problem instance. This suggests that some restrictions are essential on
populations to devise $O(\\log(1/\\delta))$ algorithms with $1 - \\delta$
correctness guarantees. We note that under restriction on population momen
ts\, such methods are easily designed. We also observe that sequential me
thods from stochastic multi-armed bandit literature can be adapted to devi
se such algorithms.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=4515
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4515
END:VEVENT
BEGIN:VEVENT
SUMMARY:Routing and Staffing when Servers are Strategic
DTSTART;VALUE=DATE-TIME:20150811T103000Z
DTEND;VALUE=DATE-TIME:20150811T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4445@cern.ch
DESCRIPTION:Traditionally\, research focusing on the design of routing and
staffing policies for service systems has modeled servers as having fixed
(possibly heterogeneous) service rates. However\, service systems are gen
erally staffed by people\, who respond to workload incentives: how hard a
person works can depend both on how much work there is\, and how the work
is divided between the people responsible for it. This observation has con
sequences when modeling performance of service systems where routing and s
taffing policies control such workload incentives\, and our objective in t
his work is to investigate those consequences. Under a simple model for st
rategic servers (that choose their service rate in order to maximize a tra
de-off between a “cost of effort” and a “value of idleness”)\, and
in the context of the M/M/N queuing model\, we characterize the symmetric
Nash equilibrium service rate under any routing policy that routes based
on the server idle time\, e.g.\, Longest-Idle-Server-First.\n\nWe find tha
t the system must operate in a Quality-Driven regime\, in which servers ha
ve idle time\, in order for an equilibrium to exist. Then\, within the cla
ss of admissible policies\, we (asymptotically) solve the problem of minim
izing the total cost\, when the staffing and waiting costs are linear. Fin
ally\, we end by exploring the question of whether routing policies that a
re based on the service rate\, instead of the server idle time\, can impro
ve system performance.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=4445
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4445
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dynamic Primal Dual Algorithms for Vertex Cover and Matching
DTSTART;VALUE=DATE-TIME:20150812T090000Z
DTEND;VALUE=DATE-TIME:20150812T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4496@cern.ch
DESCRIPTION:Consider a scenario where we are given an input graph G = (V\,
E)\, and at each time-step either a new edge is inserted into the graph o
r an already existing edge is deleted from the graph. In this dynamic sett
ing\, we want to maintain an approximately minimum vertex cover and an app
roximately maximum matching in G.\n\nOur main result is a deterministic pr
imal-dual algorithm for this problem. The algorithm maintains a (2+\\epsil
on)-approximate minimum vertex cover in O(\\log n/\\epsilon^2) amortized u
pdate time\, where n denotes the number of nodes. This answers an open que
stion from Onak and Rubinfield [STOC' 2010]. We can also maintain a (3+\\e
psilon)-approximate maximum matching in O(m^{1/3}) amortized update time\,
where m denotes the number of edges in the graph.\n\nWe also describe ext
ensions of our framework to dynamic set cover and dynamic b-matching probl
ems (joint work with Monika Henzinger and Guiseppe Italiano\, based on wor
k that appeared in SODA 2015 and ICALP 2015).\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=4496
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4496
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stable Point Processes\, Branching Random Walks and a Prediction o
f Brunet and Derrida
DTSTART;VALUE=DATE-TIME:20150814T103000Z
DTEND;VALUE=DATE-TIME:20150814T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4508@cern.ch
DESCRIPTION:Stable point processes were introduced and characterized by Da
vydov\, Molchanov and Zuyev (2008). They showed that such a point process
can always be represented as a scale mixture of iid copies of one point pr
ocess with the scaling points coming from an independent Poisson random me
asure. We obtain such a point process as the weak limit of a sequence of p
oint processes induced by a branching random walk with jointly regularly v
arying displacements. In particular\, we show that a prediction of two sta
tistical physicists\, Brunet and Derrida (2011)\, remains valid in this se
tup\, and recover a slightly improved version of a result of Durrett (1983
) (this talk is based on a joint work with Ayan Bhattacharya and Rajat Sub
hra Hazra).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=4508
LOCATION: AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4508
END:VEVENT
BEGIN:VEVENT
SUMMARY:Polynomial Approximations Over $\\mathbb{Z}/{p^k}\\mathbb{Z}$
DTSTART;VALUE=DATE-TIME:20150818T103000Z
DTEND;VALUE=DATE-TIME:20150818T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4524@cern.ch
DESCRIPTION:It is known from the works of Lovett et al.\, Tao and Ziegler\
, and more recently\, Bhowmick and Lovett\, that there are Boolean functio
ns which do not correlate well with classical polynomials of a certain deg
ree but have good correlation with some non-classical polynomial of the sa
me degree.\n\nNoting that the notion of approximation is different from th
at of correlation in the case of non-classical polynomials\, Bhowmick and
Lovett asked the following questions:\n\\begin{itemize}\n\\item Do non-cla
ssical polynomials of degree $o(\\sqrt{n})$ approximate the majority funct
ion better than classical polynomials of the same degree?\n\\item Is there
any Boolean function for which non-classical polynomials offer an advanta
ge over classical polynomials in the case of approximation?\n\\end{itemize
}\n\nWe give a negative answer to the first question. We do so by studying
polynomials over rings of the form $\\mathbb{Z}/{p^k}\\mathbb{Z}$ and obs
erving that non-classical polynomial are a special case of such polynomial
s. Our proof essentially involves proving tight bounds for \\textit{weak r
epresentations} of the majority function over $\\Ringm{p^k}$\, strengtheni
ng classical results of Szegedy and Smolensky.\n\nFor the second question\
, we give a positive answer by showing that elementary symmetric polynomia
ls of a suitable degree are well approximated by non-classical polynomials
(this is joint work with Prahladh Harsha and Srikanth Srinivasan).\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4524
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4524
END:VEVENT
BEGIN:VEVENT
SUMMARY:Graph Colouring
DTSTART;VALUE=DATE-TIME:20150821T103000Z
DTEND;VALUE=DATE-TIME:20150821T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4530@cern.ch
DESCRIPTION:Given a graph $G$\, a valid colouring of $G$ is defined as an
assignment of colours to the vertices of $G$ such that no two adjacent ver
tices share the same colour. If it is possible to produce a valid colourin
g of $G$ using at most $k$ different colours\, then we say that $G$ is $k$
-colourable. It is well-known and easy to see that every graph with maximu
m degree $d$ is $(d+1)$-colourable. The complete graph and the odd cycle a
re two trivial examples of graphs that are $(d+1)$-colourable but not $d$-
colourable.\n\nIn 1941\, Leonard Brooks proved that these are\, in fact\,
the only two cases where $d+1$ colours are necessary. In other words\, if
$G$ is not an odd cycle and does not contain $K_{d+1}$ (the complete graph
on $d+1$ vertices) as a subgraph\, then $d$ colours are sufficient to pro
duce a valid colouring of $G$. In 1973\, Lovasz came up with a nice one pa
ge proof of Brooks' theorem. We will discuss this proof in the talk. Time
permitting\, we will also see whether a triangle free graph is $c$-coloura
ble\, for some constant $c$.\n\nhttps://indico.tifr.res.in/indico/conferen
ceDisplay.py?confId=4530
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4530
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Deterministic Polynomial Time Algorithm for Word Problem Over th
e Free Skew Field
DTSTART;VALUE=DATE-TIME:20150825T103000Z
DTEND;VALUE=DATE-TIME:20150825T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4500@cern.ch
DESCRIPTION:We study the word problem for the free skew field of non-commu
tative rational functions. We prove that an existing algorithm due to Gurv
its is actually a deterministic polynomial time algorithm for this problem
(over the rationals). Our analysis is simple\, providing explicit bounds
on the ``capacity'' measure of totally positive operators introducted by G
urvits.\n\nThis problem has a rich history and has appeared in various for
ms in a remarkable number of mathematical and computational areas. Besides
non-commutative algebra\, it also has various connections to computationa
l complexity and de-randomization\, commutative invariant theory\, quantum
information theory\, system theory\, automata and language theory. I plan
to explain some of these connections\, as well as the central open proble
m relating them - the ``fullness dimension''.\n\nNo special background wil
l be assumed\, as it can be\, the problem\, algorithm and analysis can be
framed in the language of linear algebra (this is joint work with Rafael O
liveira and Avi Wigderson).\n\nhttps://indico.tifr.res.in/indico/conferenc
eDisplay.py?confId=4500
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4500
END:VEVENT
BEGIN:VEVENT
SUMMARY:Statistical Model Checking for Unbounded Temporal Properties
DTSTART;VALUE=DATE-TIME:20150901T100000Z
DTEND;VALUE=DATE-TIME:20150901T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4537@cern.ch
DESCRIPTION:In this talk\, I present a new algorithm for the statistical m
odel checking of Markov chains with respect to unbounded temporal properti
es\, such as reachability and full linear temporal logic. The main idea is
that we monitor each simulation run on the fly\, in order to detect quick
ly if a bottom strongly connected component is entered with high probabili
ty\, in which case the simulation run can be terminated early. As a result
\, our simulation runs are often much shorter than required by termination
bounds that are computed a priori for a desired level of confidence and s
ize of the state space. In comparison to previous algorithms for statistic
al model checking\, for a given level of confidence\, our method is not on
ly faster in many cases but also requires less information about the syste
m\, namely\, only the minimum transition probability that occurs in the Ma
rkov chain. In addition\, our method can be generalised to unbounded quant
itative properties such as mean-payoff bounds.\n\nhttps://indico.tifr.res.
in/indico/conferenceDisplay.py?confId=4537
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4537
END:VEVENT
BEGIN:VEVENT
SUMMARY:Drainage Networks and the Brownian Web
DTSTART;VALUE=DATE-TIME:20150908T103000Z
DTEND;VALUE=DATE-TIME:20150908T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4501@cern.ch
DESCRIPTION:River basin geomorphology is a very old subject of study initi
ated by Horton (1945). Various statistical models of drainage networks hav
e been proposed. Each such model is a random directed graph with its own n
uances.\n\nIn recent years physicists have been interested in these models
because of the commonality of such branching networks in many areas of st
atistical physics (see Rodriguez-Iturbe and Rinaldo (1997) for a detailed
survey).\n\nWe discuss the geometric features of one such model and also
its scaling limit. The scaling limit of this model is the Brownian web\, w
hich has lately been the focus of extensive study among probabilists.\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4501
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4501
END:VEVENT
BEGIN:VEVENT
SUMMARY:Algebraic Methods in Distributed Graph Algorithms
DTSTART;VALUE=DATE-TIME:20150915T103000Z
DTEND;VALUE=DATE-TIME:20150915T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4560@cern.ch
DESCRIPTION:We consider a computer network\, represented by a communicatio
n graph: each node represents a processor and the communication is done sy
nchronously along the edges. Can such a network compute its own graph's pr
operties? E.g.\, how hard is it to list all the triangles in the graph\, t
o find its shortest cycle\, or to compute all-pairs shortest paths?\n\nWe
will give an introduction to the relevant computation models. Then\, we wi
ll discuss some recent results showing how to solve these problems faster
using algebraic techniques (based on a joint work with Keren Censor-Hillel
\, Petteri Kaski\, Janne H. Korhonen\, Christoph Lenzen and Jukka Suomela)
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4560
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4560
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hashing Based Almost-uniform Generation and Model Counting
DTSTART;VALUE=DATE-TIME:20150922T103000Z
DTEND;VALUE=DATE-TIME:20150922T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4570@cern.ch
DESCRIPTION:Given a propositional logic formula $F$ in $CNF$\, we consider
two related problems: (i) generating satisfying assignments of F uniforml
y at random\, and (ii) counting the number of satisfying assignments. If w
e could list all satisfying assignments of F explicitly\, both problems ar
e trivial. In practical applications\, however\, it is infeasible to list
all satisfying assignments of $F$. Therefore\, the above problems need to
be solved without generating all satisfying assignments in the first place
. Both problems have been of significant interest in the theoretical compu
ter science community and in more applied areas. For instance\, propositio
nal model-counting is the canonical $#P$-complete problem\, strong theoret
ical results have been demonstrated in the past on the connection between
almost uniform generation and approximate model counting. Similarly\, sol
utions to the above problems find many applications in verification\, test
ing and probabilistic inference and analysis. Earlier approaches to these
problems belong to one of two distinct categories: (i) Theoretically motiv
ated approaches that provide strong guarantees but do not lend themselves
to implementations that scale to practical problem sizes\, or (ii) Practic
ally motivated approaches that scale to large problem sizes but provide ve
ry weak guarantees.\n\nIn this talk\, we describe our ongoing work on brid
ging these extremes. Specifically\, we show that it is possible to design
randomized algorithms that have strong theoretical guarantees (similar to
the best known guarantees from theoretically motivated approaches) and als
o scale to practical problem sizes (comparable to those achieved by practi
cally motivated approaches). Our approach makes crucial use of universal h
ash functions\, and expose possibilities of hitherto unexplored theoretica
l connections between approximate model counting and weaker versions of un
iform witness generation.\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=4570
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4570
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimal Offline and Competitive Online Strategies for Transmitter-
Receiver Energy Harvesting
DTSTART;VALUE=DATE-TIME:20151006T103000Z
DTEND;VALUE=DATE-TIME:20151006T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4593@cern.ch
DESCRIPTION:A joint transmitter-receiver energy harvesting model is consid
ered\, where both the transmitter and receiver are powered by (renewable)
energy harvesting source. EH gives rise to instantaneous energy neutrality
constraint\, i.e. energy consumed till any time cannot be more than the t
otal accumulated energy. Assuming no information about future energy arriv
als\, the goal is to find an optimal 'online algorithm'. In this work\, bo
th optimal offline and optimal online policies are derived. The optimal on
line policy is shown to be two-competitive in the arbitrary input case.\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4593
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4593
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Bit-Probe Complexity of Set Membership
DTSTART;VALUE=DATE-TIME:20151008T090000Z
DTEND;VALUE=DATE-TIME:20151008T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4592@cern.ch
DESCRIPTION:We will consider the fundamental trade-off between the compact
ness of information representation and the efficiency of information extra
ction in the context of the set membership problem. In the set membership
problem\, a set S of size at most $n$ from a universe of size $m$ is to be
represented as a short bit vector in order to answer membership queries o
f the form ``Is x in S?'' by probing the bit vector at $t$ places. The bit
-probe complexity $s(m\,n\,t)$ is the minimum number of bits of storage n
eeded for such a scheme. Improving on the works of Buhrman\, Milterson\, R
adhakrishnan and Venkatesh (2002) and Alon and Feige (2009) we provide bet
ter bounds on $s(m\,n\,t)$ for a range of $m$\, $n$ and $t$.\n\nFor two pr
obes:\n\n $s(m\,n\,2) =O(m^{1-\\frac{1}{4n+1}})$\; this improves on a
result of Alon and Feige that states that for $n \\leq \\lg m$\, $s(m\,n\,
2) = O( m n \\lg ((\\lg m) / n) / \\lg m)$\n $s(m\,n\,2)= \\Omega(m^{1-
\\frac{1}{\\lfloor n/4 \\rfloor}})$\; in particular\, $s(m\,n\,2)=\\Omega(
m)$ for $n \\geq 4\\lg m$.\n\nFor three probes:\n\n $s(m\,n\,3) = \\til
de{O}(\\sqrt{mn}).$ This improves a result of Alon and Feige that states t
hat $s(m\,n\,2)=O(m^{\\frac{2}{3}} n^{\\frac{1}{3}})$. For $n\\approx \\l
g m$\, our general lower bound implies a lower bound of $\\Omega(\\sqrt{m}
)$ which comes close to the upper bound.\n The best non-adaptive scheme
s (when probes are made in parallel) for $t\\geq 4$ use majority decoding
to answer membership queries. For three non-adaptive probes\, we show a lo
wer bound for majority decoding schemes: $s(m\,n\,t)=\\Omega(m^{1-\\frac 1
{cn}})$ for some $c>0$. Lower bounds are also obtained for other function
s.\n\nIn general: We exhibit non-adaptive and adaptive schemes for odd $t\
\geq 5$ that use significantly less space than the schemes of Buhrman et a
l. We also obtain slight improvement over their general lower bound.\n\n(T
his is joint work with Jaikumar Radhakrishnan.)\n\nhttps://indico.tifr.res
.in/indico/conferenceDisplay.py?confId=4592
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4592
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fast Provable Non-convex Algorithms for Matrix Decomposition
DTSTART;VALUE=DATE-TIME:20151023T053000Z
DTEND;VALUE=DATE-TIME:20151023T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4602@cern.ch
DESCRIPTION:Matrix decomposition\, where we wish to approximate a given ma
trix as a sum or product of two or more structured matrices\, plays a fund
amental role in many machine learning applications. Principal components a
nalysis (PCA) and k-means clustering\, which are ubiquitous in machine lea
rning\, are for instance specializations of this problem. For many such pr
oblems\, there is a dichotomy between algorithms that have provable guaran
tees and those that are used in practice. The ones with provable guarantee
s\, though polynomial time\, are time and memory intensive where as the on
es that are used in practice are not well understood but turn out to be ve
ry efficient and have good empirical performance. In this talk\, we focus
on one such empirically popular heuristic called alternating minimization
(AltMin). Given a function that we wish to minimize\, AltMin divides the v
ariables into two or more sets and sequentially optimizes over one of them
while holding the rest fixed. In many settings\, it turns out that though
minimizing the function jointly over all variables is computationally har
d\, optimizing over any one set of variables can be done efficiently. We p
rovide the first performance guarantees for AltMin for the problems of\n\n
1. Matrix Completion\, where we wish to recover a low rank matrix from par
tial observations of some of the entries\, and\n2. Robust PCA\, where we w
ish to decompose a given matrix as the sum of a low rank matrix and a spar
se matrix.\n\nFor both of these problems\, our results give faster algorit
hms than those known before. More broadly\, our results indicate the poten
tial use of non-convex optimization to design faster algorithms. I will me
ntion some ongoing work and future research avenues in this direction.\n\n
Bio:Praneeth Netrapalli is currently a postdoctoral researcher at Microsof
t Research New England in Cambridge MA. He obtained his MS and PhD from UT
Austin and B-Tech from IIT Bombay\, all in Electrical engineering. Before
pursuing his PhD\, he spent two years at Goldman Sachs\, Bangalore as a q
uantitative analyst where he worked on pricing derivatives. His research f
ocuses on designing efficient and provable algorithms for machine learning
problems.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=4602
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4602
END:VEVENT
BEGIN:VEVENT
SUMMARY:Modeling and Constructing Cryptography for the Cloud
DTSTART;VALUE=DATE-TIME:20151105T050000Z
DTEND;VALUE=DATE-TIME:20151105T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4614@cern.ch
DESCRIPTION:We live in technologically exciting times. The big data revolu
tion promises technology bordering on science fiction -- personalized medi
cine via genome sequencing\, the "internet of things" offering the potent
ial of smart cities\, smart appliances for healthcare\, smart metering and
many such applications. But as our technology has increased\, so has our
vulnerability. In the age of Wikileaks and Snowden\, we worry increasingly
about the security of our data and about our privacy at large. Can we ena
ble technology while still having reasonable guarantees on security? This
is a complex question and we study how cryptography can offer solutions to
the demands placed by the above applications.\n\nBroadly speaking\, the c
hallenges placed on cryptography are twofold.\n\nModelling: The usage scen
arios in which cryptographic schemes are deployed are becoming more and mo
re complex\, hence modeling security mathematically and designing systems
to achieve it is becoming increasingly harder. Even basic cryptographic ob
jects like public key encryption and signatures face new modeling and achi
evability challenges.\n\nConstructions: The functionality being asked from
cryptographic objects is becoming elaborate and sophisticated\, with many
nontrivial generalizations of standard primitives\, like encryption and s
ignatures\, being desired.\n\nIn this talk\, I will discuss some recent re
sults pertaining to both these aspects.\n\nFirst\, I will describe a new m
odel of cryptographic computation\, that unifies and extends existing cryp
tographic primitives such as Obfuscation\, Functional Encryption\, Fully H
omomorphic Encryption\, Witness encryption\, Property Preserving Encryptio
n and the like\, all of which can be cleanly modeled in our framework. We
provide a new definition of security that finds the sweet spot between ach
ievability and impossibility -- implying most achievable security definiti
ons while sidestepping the impossibilities that plague "too-strong" defini
tions. We also provide powerful reduction and composition theorems that yi
eld a modular means to build and analyze complicated cryptographic objects
from simpler ones.\n\nSecond\, I will describe some recent work in the co
nstruction of novel encryption schemes that generalize and extend public k
ey encryption. An attractive feature of one of our constructions is an "on
line-offline" property\, that enables the encryptor to do a large amount o
f work before it receives the data\, i.e. "offline"\, so that encryption i
s very efficient after data becomes available. This makes the construction
suitable for streaming data applications.\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=4614
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4614
END:VEVENT
BEGIN:VEVENT
SUMMARY:Algorithms\, Games and Evolution
DTSTART;VALUE=DATE-TIME:20151116T060000Z
DTEND;VALUE=DATE-TIME:20151116T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4625@cern.ch
DESCRIPTION:Even the most seasoned students of evolution\, starting with D
arwin himself\, have occasionally expressed amazement that the mechanism o
f natural selection has produced all of life as we see it around us. Or st
ated from a computational perspective\, what algorithm could possibly achi
eve Nature's solution to the problems of robotics\, vision and theorem pro
ving in about 10^12 generations (steps)? We demonstrate that in the regime
of weak selection\, the standard equations of population genetics describ
ing natural selection in the presence of recombination become identical to
those of a repeated game between genes played according to multiplicative
weight updates (MWUA)\, an algorithm known to be surprisingly powerful an
d versatile. A dual description of MWUA shows that it maximizes a trade-of
f between cumulative performance and entropy\, which suggests a new view o
n the maintenance of diversity in evolution (based on joint work with Eric
k Chastain\, Adi Livnat and Christos Papadimitriou).\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=4625
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4625
END:VEVENT
BEGIN:VEVENT
SUMMARY:Primal-Dual Algorithms in Scheduling
DTSTART;VALUE=DATE-TIME:20151117T103000Z
DTEND;VALUE=DATE-TIME:20151117T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4649@cern.ch
DESCRIPTION:Recent years have seen application of Linear programming techn
iques to solve problems in scheduling. One technique that has been applied
very effectively to both online and offline scheduling problems is the Pr
imal-Dual method. In this talk I will take two examples from our own resea
rch to illustrate some of the key ideas. The problems I will consider are:
\n\n- the online problem of minimizing flow time on unrelated machines\n\n
- the offline problem of minimizing weighted flow time on a single machine
.\n\nThe talk will assume only basic familiarity with linear programming a
nd duality.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=4649
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4649
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lower Bounds for Shallow Circuits
DTSTART;VALUE=DATE-TIME:20151123T103000Z
DTEND;VALUE=DATE-TIME:20151123T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4672@cern.ch
DESCRIPTION:Over the last 3--4 years\, there has been a sudden surge of re
sults in arithmetic circuit lower bounds\, specifically for shallow circui
ts (depth 3 or 4). The main motivation for studying depth-4 circuits is a
result of Agrawal and Vinay (and subsequent strengthening by Koiran and Ta
venas) that states that any arithmetic circuit of size s that computes an
n-variate degree d polynomial can be simulated by a structured homogeneous
depth-4 circuit of size n^{Omega (sqrt{d})}. Flipping this around\, this
implies that if we can find an explicit polynomial that requires such stru
ctured depth-4 circuits of size n^{omega(sqrt {d})}\, then we would have p
roved lower bounds for general arithmetic circuits!\n\nIn 2013\, in a join
t work with Gupta\, Kamath and Kayal\, we showed a lower bound of 2^{Omega
(sqrt{d})} for the dxd permanent or determinant using a technique called
'shifted partial derivatives' (introduced by [Kayal]). Subsequent to this
result\, there has been a sequence of improvements of various sorts (impro
ving the lower bound\, or dealing with a larger circuit class etc.) thereb
y improving our understanding of depth four circuits. The current state of
the art is an n^{Omega(sqrt{d})} lower bound for the class of homogeneous
depth-4circuits with no structural restrictions [Kayal-Limaye-Saha-Sriniv
asan\, Kumar-Saraf].\n\nIn this talk\, I shall give a brief overview of th
e progress made in the last 3 years and put the various results in context
. Then\, I shall present a recent work with Mrinal Kumar that gives an exp
onential lower bound for depth-5 circuits over small finite fields that co
mbines the techniques developed so far with an old result of Grigoriev-Kar
pinski for depth-3 lower bounds.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=4672
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4672
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dimension Reduction for Efficient Algorithms and Visualization
DTSTART;VALUE=DATE-TIME:20151130T103000Z
DTEND;VALUE=DATE-TIME:20151130T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4685@cern.ch
DESCRIPTION:High dimensional problems are ubiquitous in machine learning a
nd many other fields such as biology\, finance\, environmental science\, m
edicine\, etc. For example\, a 5 mega pixel image can be thought as a vect
or in 5 million dimensional vector space\, similarly a document can also b
e considered as a vector indexed by words in high dimensional vector space
. Dimension reduction is a well studied problem in machine learning. The m
ain reason behind studying this has been three folds: i) to avoid the curs
e of dimensionality\, ii) to develop efficient algorithms\, and iii) to de
velop methods for visualization of high dimensional data. In this talk we
will discuss three problems about dimension reduction where the focus of t
he first one is on classification and the other two is on visualizing high
dimensional data.\n\n1. In the first part we will discuss how mutual info
rmation can be used to reduce the output dimensions.\n\n2. We study a gene
ralization of the well known data visualization technique t-distributed st
ochastic neighbor embedding (t-SNE).\n\n3. We present a new algorithm for
embedding large dimensional data into two or three dimensions for visualiz
ation.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=46
85
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4685
END:VEVENT
BEGIN:VEVENT
SUMMARY:How Well Can Graphs Represent Wireless Interference?
DTSTART;VALUE=DATE-TIME:20151203T060000Z
DTEND;VALUE=DATE-TIME:20151203T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4689@cern.ch
DESCRIPTION:From the early days of wireless networking research\, graphs h
ave commonly been used to model wireless communications and interference.
Much theoretical work was done on simple range models\, like Unit Disc Gra
phs. Unfortunately\, such models were generally found to be quite far from
the reality on the ground. This has led to sustained efforts in recent ye
ars to formulate and analyze algorithms in the SINR or physical model. Dea
ling with graphs is\, however\, more preferable\, as it makes all analysis
easier and allows for the transfer of a lot of well-explored theory. We r
e-examine how well graphs can capture wireless receptions as encoded in SI
NR relationships\, placing them in a framework in order to understand the
limits of such modelling.\n\nWe propose a new type of conflict graphs and
show that they come very close to capturing the SINR interference relation
ships. This leads to greatly improved approximation algorithm for a host o
f fundamental problems\, including link scheduling and packet scheduling.
The performance guarantees obtained are O(log* Delta)\, where Delta is the
ratio between the longest and the shortest link length\, improving on pre
vious logarithmic factors.\n\nWe show that this is tight: no conflict grap
h representation schema can guarantee a better than Omega(log* Delta)-appr
oximation. We can view this a tight characterization on the "price" of usi
ng the simpler graph-based representation in the setting of wireless commu
nication (this is joint work with Tigran Tonoyan).\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=4689
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4689
END:VEVENT
BEGIN:VEVENT
SUMMARY:Secret Key Agreement: General Capacity and Second-Order Asymptotic
s
DTSTART;VALUE=DATE-TIME:20151204T053000Z
DTEND;VALUE=DATE-TIME:20151204T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4671@cern.ch
DESCRIPTION:We revisit the problem of secret key agreement using interacti
ve public communication for two parties and propose a new secret key agree
ment protocol. The protocol attains the secret key capacity for general ob
servations and attains the second-order asymptotic term in the maximum len
gth of a secret key for independent and identically distributed observatio
ns. In contrast to the previously suggested secret key agreement protocols
\, the proposed protocol uses interactive communication. In fact\, the sta
ndard one-way communication protocol used prior to this work fails to atta
in the asymptotic results above. Our converse proofs rely on a recently es
tablished upper bound for secret key lengths. Both our lower and upper bou
nds are derived in a single-shot setup and the asymptotic results are obta
ined as corollaries (it is a joint work with M. Hayashi and H. Tyagi (arXi
v:1411.0735).\n\nBio: Shun Watanabe received B.E.\, M.E.\, and Ph.D. degre
e from Tokyo Institute of Technology in 2005\, 2007\, and 2009\, respectiv
ely. During April 2009 to February 2015\, he was an assistant professor of
the Department of Information Science and Intelligence Systems at the Uni
versity of Tokushima. During April 2013 to March 2015\, he was a visiting
assistant professor of the Institute for Systems Research at the Universi
ty of Maryland\, College Park. Since February 2015\, he has been a tenure
track associate professor of the Department of Computer and Information Sc
iences at Tokyo University of Agriculture and Technology.\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=4671
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4671
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximation Algorithms for Geometric Covering and Packing Proble
ms
DTSTART;VALUE=DATE-TIME:20151208T103000Z
DTEND;VALUE=DATE-TIME:20151208T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4696@cern.ch
DESCRIPTION:In the classic set cover problem we are given a finite ground
set $X$ of elements and a collection of subsets of $X$ whose union covers
$X$. The problem is to find a sub-collection of smallest cardinality from
the given collection that still forms a cover of $X$. In the classic packi
ng problem\, we want the largest cardinality sub-collection that are pairw
ise disjoint. These classic problems are well reasonably well understood i
n the abstract setting.\n\nIn this talk\, we consider these problems in th
e geometric setting. Here\, the elements of $X$ are points in the plane\,
and the subsets of $X$ are defined by geometric regions. These problems ar
e not as well understood in the geometric setting. For example\, there are
gaps between the unweighted setting described above\, and the weighted se
tting in which the sets have weights and we want to find the minimum (or m
aximum) weight sub-collection that forms a cover (or forms a packing).\n\n
We present approximation algorithms for the covering problem where the set
s are non-piercing\, i.e.\, for any two regions simply connected $A$ and $
B$\, the regions $AB$ and $BA$ are connected. We extend these results to a
related covering problem\, namely the dominating set problem as well as a
generalization of the packing problem where the points (or sets) have cap
acities that are bounded by a constant. The results in this talk were obta
ined jointly with Aniket Basu\, Sathish Govindarajan\, Nabil Mustafa and S
aurabh Ray.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=4696
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4696
END:VEVENT
BEGIN:VEVENT
SUMMARY:Interpolation Synthesis for Quadratic Polynomial Inequalities and
Combination with Theory of Equality with Uninterpreted Function Symbols (E
UF)
DTSTART;VALUE=DATE-TIME:20151222T103000Z
DTEND;VALUE=DATE-TIME:20151222T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4727@cern.ch
DESCRIPTION:An algorithm for generating interpolants for formulas which ar
e conjunctions of quadratic polynomial inequalities (both strict and nonst
rict) is proposed. The algorithm is based on a key observation that quadra
tic polynomial inequalities can be linearized if they are concave. A gener
alization of Motzkin's transposition theorem is proved\, which is used to
generate an interpolant between two mutually contradictory conjunctions of
polynomial inequalities\, in a way similar to the linear inequalities cas
e. This can be done efficiently using semi-definite programming but forsa
king completeness. A combination algorithm is given for the combined theo
ry of concave quadratic polynomial inequalities and the equality theory ov
er uninterpreted functions symbols using a hierarchical framework for comb
ining interpolation algorithms for quantifier-free theories. A preliminary
implementation has been explored.\n\nhttps://indico.tifr.res.in/indico/co
nferenceDisplay.py?confId=4727
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4727
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hardness of Approximate Coloring
DTSTART;VALUE=DATE-TIME:20160108T083000Z
DTEND;VALUE=DATE-TIME:20160108T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4753@cern.ch
DESCRIPTION:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
4753
LOCATION: AG-69
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4753
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Introduction to Epistemic Logics
DTSTART;VALUE=DATE-TIME:20160108T053000Z
DTEND;VALUE=DATE-TIME:20160108T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4769@cern.ch
DESCRIPTION:Epistemic logics are logics that attempt to model knowledge. I
n this talk\, I will give an introduction to epistemic logic and focus on
the logic S5\, adding common knowledge and dynamic modalities.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=4769
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4769
END:VEVENT
BEGIN:VEVENT
SUMMARY:Distance Preserving Minors in Interval Graphs
DTSTART;VALUE=DATE-TIME:20160109T043000Z
DTEND;VALUE=DATE-TIME:20160109T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4770@cern.ch
DESCRIPTION:We consider undirected\, connected graphs with nonnegative wei
ghts on the edges. Additionally\, a special subset of vertices called term
inals is provided as input. A graph $H$ is said to be a distance preservin
g minor of $G$ if: (i) $H$ is a minor of $G$\, and (ii) the distance betwe
en each pair of terminals is exactly the same in $G$ and $H$. Note that th
e edge weights can be reassigned in $H$ (as long as they are nonnegative).
Given a family of graphs $\\mathcal{F}$\, let $f(k\,\\mathcal{F})$ be the
minimum integer such that every graph in $\\mathcal{F}$ with $k$ terminal
s admits a distance preserving minor with at most $f(k\,\\mathcal{F})$ ver
tices.\n\nIn this work\, we explore interval graphs. Given an interval gra
ph $G$ on $k$ terminals\, we describe a method of producing a distance pre
serving minor of $G$ having at most $k^2$ vertices. Thus\, $f(k\,\\mathcal
{I})\\le k^2$\, where $\\mathcal{I}$ is the family of all interval graphs.
We also show that there are interval graphs on which our method does not
perform optimally. In particular\, there exists an interval graph which ad
mits a distance preserving minor of size $O(k\\log k)$\, but our method pr
oduces a distance preserving minor of size $\\Omega(k^2)$ for it. Thus\, t
he problem of finding an optimal distance preserving minor for interval gr
aphs remains open.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=4770
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4770
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Theory of Algorithmic Improvisation
DTSTART;VALUE=DATE-TIME:20160111T053000Z
DTEND;VALUE=DATE-TIME:20160111T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4771@cern.ch
DESCRIPTION:Improvisation is often described as "acting without preparatio
n"\, with an element of randomness. Can an algorithm improvise? What does
it mean for an algorithm to improvise? What applications can benefit from
algorithmic improvisation?\n\nIn this talk\, I will explore these question
s by presenting "control improvisation"\, a formalization of the notion of
algorithmic improvisation. A simple form of control improvisation is to f
ind a random generator of strings from a formal language satisfying certai
n hard\, soft\, and randomness requirements. Applications of control impro
visation are surprisingly diverse\, ranging from robotics to software test
ing to computational music. I will present the theory of control improvisa
tion\, an analysis of its computational complexity\, and some preliminary
applications that we have explored at Berkeley.\n\nBio: Sanjit A. Seshia i
s an Associate Professor in the Department of Electrical Engineering and C
omputer Sciences at the University of California\, Berkeley. He received a
n M.S. and Ph.D. in Computer Science from Carnegie Mellon University\, and
a B.Tech. in Computer Science and Engineering from the Indian Institute o
f Technology\, Bombay. His research interests are in dependable computing
and computational logic\, with a current focus on applying automated forma
l methods to problems in cyber-physical systems\, computer security\, elec
tronic design automation\, and synthetic biology. His Ph.D. thesis work on
the UCLID verifier and decision procedure helped pioneer the area of sati
sfiability modulo theories (SMT) and SMT-based verification. He is co-auth
or of a textbook on embedded systems used in more than 50 countries and ha
s led the development of technologies for cyber-physical systems education
based on formal methods. His awards and honors include a Presidential Ear
ly Career Award for Scientists and Engineers (PECASE) from the White House
\, an Alfred P. Sloan Research Fellowship\, and the School of Computer Sci
ence Distinguished Dissertation Award at Carnegie Mellon University.\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4771
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4771
END:VEVENT
BEGIN:VEVENT
SUMMARY:Group Knowledge\, Signals and Behavior
DTSTART;VALUE=DATE-TIME:20160112T103000Z
DTEND;VALUE=DATE-TIME:20160112T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4764@cern.ch
DESCRIPTION:We discuss two issues regarding group knowledge.\n\n1) How can
we tell by the behavior of agents in a group what their beliefs are or at
least what they are not? For instance someone going out without an umbr
ella reveals the fact that she does not know (believe) it is raining. She
does not need to tell us\, we can infer what she believes from what she d
oes. We deal not only with beliefs of agents about the world but also wit
h their beliefs about the beliefs of other agents and so on.\n\nA principa
l tool used is the notion of rationalizable strategy\, defined by Bernheim
and Pearce in Econometrica 1984 but generalized to take into account the
different states of knowledge of the agents. The tools developed apply n
ot only to agents who are competent users of language but also animals and
children and agents who have reason to deceive.\n\n2) How can we influenc
e the behavior of agents by influencing what they know or what they believ
e? We give a theoretical account of this and also show that for any stat
e of knowledge described by a finite Kripke structure\, there is an n-tupl
e of signals that can be sent\, one to each agent which will create precis
ely that state of knowledge. We distinguish between cautious and aggressi
ve agents and point out that given the same choice situation\, the two kin
ds of agents may act differently. Thus knowing the temperament of the age
nt helps us to predict its behavior.\n\nThe work in 1) is work under progr
ess. Work in 2) is joint with Cagil Tasdemir\, a recent doctorate from
CUNY\, and Andreas Witzel\, now working at Google in New York.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=4764
LOCATION: A-212 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4764
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Complexity of Expansion Problems
DTSTART;VALUE=DATE-TIME:20160119T060000Z
DTEND;VALUE=DATE-TIME:20160119T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4749@cern.ch
DESCRIPTION:Graph-partitioning problems are a central topic of research in
the study of algorithms and complexity theory. They are of interest to th
eoreticians with connections to error correcting codes\, sampling algorith
ms\, metric embeddings\, among others\, and to practitioners\, as algorith
ms for graph partitioning can be used as fundamental building blocks in ma
ny applications. One of the central problems studied in this field is the
sparsest cut problem\, where we want to compute the cut which has the leas
t ratio of number of edges cut to size of smaller side of the cut. This ra
tio is known as the expansion of the cut. \n\nIn this talk\, I will discus
s three notions of expansion - edge expansion in graphs\, vertex expansion
in graphs and edge expansion in hypergraphs. I will define suitable notio
ns of spectra and show how the notion of the celebrated "Cheeger's Inequal
ity" goes across these three problems. I will also talk about higher order
variants of these notions of expansion (i.e. notions of expansion corresp
onding to partitioning the graph/hypergraph into more than two pieces\, et
c.)\, and also about approximation algorithms for these problems.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4749
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4749
END:VEVENT
BEGIN:VEVENT
SUMMARY:System Utility Maximisation With Fast Network and User Adaptation
DTSTART;VALUE=DATE-TIME:20160119T100000Z
DTEND;VALUE=DATE-TIME:20160119T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4775@cern.ch
DESCRIPTION:Kelly\, Maulloo\, and Tan (1998) proposed a decomposition of a
system utility maximisation problem into a network utility maximisation p
roblem\, to be solved by a network entity\, and a set of decoupled user op
timisation problems\, to be solved by the individual users. They proposed
an iterative scheme where the users signal their willingness to pay (a sca
lar quantity) while the network signals the charges per unit flow. Their s
cheme involves a slow adaptation by the network and a fast adaptation by t
he users. Their scheme converges to the system optimal solution and has th
e advantage that the network entity need not know the users' utility funct
ions and the individual users need not know the network state or utility f
unctions of other users. In this talk\, we will discuss a special case and
a modified algorithm where the network also makes a fast adapation. We wi
ll then show that the algorithm converges to the global optimum solution.\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4775
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4775
END:VEVENT
BEGIN:VEVENT
SUMMARY:Towards Efficient Counter Example to Additivity of Minimum Output
Renyi $p$ Entropy
DTSTART;VALUE=DATE-TIME:20160120T093000Z
DTEND;VALUE=DATE-TIME:20160120T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4796@cern.ch
DESCRIPTION:Research in quantum information theory suggests that if a quan
tum channel is used to transmit classical information then the capacity (a
lso referred to as classical capacity) of a quantum channel is super addit
ive. This has evolved as a result of counter examples to additivity conjec
ture of quantum information theoretic quantities and their relation with o
ne other.\n\nIn this talk\, we discuss the counter example to additivity c
onjecture of minimal output Renyi p entropy of a quantum channel\, which w
as given by Hayden and Winter. But the technique used by them doesn't give
an efficient implementation of the underlying quantum channel\, in the se
nse that their analysis is based on selecting channel unitaries at random
according to Haar measure\, which cannot be constructed efficiently.\n\nWe
discuss the same counter example but the channel unitaries are now chosen
from a finite size approximate t design which have efficient implementati
ons. This is based on a geometric functional analytic result called Dvoret
zky's theorem\, as the additivity conjecture is equivalent to the multipli
cativity of super operator norm which has a functional analytic form.\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4796
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4796
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Attempt to Derandomize the Random Unitary Matrices Used for Dec
oupling in Quantum Information Theory
DTSTART;VALUE=DATE-TIME:20160120T083000Z
DTEND;VALUE=DATE-TIME:20160120T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4802@cern.ch
DESCRIPTION:The correlation of a quantum system A to an initially correlat
ed system E generally decreases when A undergoes an evolution separated f
rom E. Here we study the conditions under which the correlation disappears
completely\, resulting in a decoupling of A from E. Decoupling is a gener
al way to prove many coding theorems in quantum information theory. Howeve
r existing decoupling constructions use random unitaries which are ineffic
ient to implement. Hence the quest for derandomization.The talk shall desc
ribe the decoupling theorem and time permitting\, sketch our attempt to de
randomize it using a mathematical tool called unitary t-designs.\n\nhttps:
//indico.tifr.res.in/indico/conferenceDisplay.py?confId=4802
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4802
END:VEVENT
BEGIN:VEVENT
SUMMARY:Graphical Models: Algorithms and Complexity
DTSTART;VALUE=DATE-TIME:20160122T060000Z
DTEND;VALUE=DATE-TIME:20160122T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4778@cern.ch
DESCRIPTION:Probabilistic graphical models provide a very useful framework
for studying several problems across theoretical computer science and sta
tistics. Undirected graphical models have also been studied extensively in
statistical mechanics under the name of "spin systems".\n\nThe first part
of the talk illustrates two examples of the crucial role played by ideas
from statistical mechanics in the study of several natural algorithmic pro
blems in theoretical computer science and combinatorics via this connectio
n to spin systems. The first connection is between a novel extension of th
e Lee-Yang theorem and the computational complexity of computing averages
such as the mean magnetization of the Ising model and the average size of
matchings. The second example illustrates the interplay between the study
of the correlation decay phenomenon and the problems of approximate counti
ng and sampling. In both cases\, the algorithmic view also contributes bac
k to the study of spin systems.\n\nThe second part of the talk considers c
ausal inference in directed graphical models. In particular\, we study th
e "condition number" of the causal inference problem\, and show that there
are graphical models in which the problem is extremely sensitive to error
s in the input. We then and propose several future directions for combatin
g this ill-conditioning.\n\nThe results in this talk are based on collabor
ations with Leonard J. Schulman\, Alistair Sinclair\, Daniel Štefankovič
\, and Yitong Yin.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=4778
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4778
END:VEVENT
BEGIN:VEVENT
SUMMARY:Transcendental Numbers and Liouville's Theorem
DTSTART;VALUE=DATE-TIME:20160129T103000Z
DTEND;VALUE=DATE-TIME:20160129T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4834@cern.ch
DESCRIPTION:In 1853\, Liouville proved a fundamental result concerning app
roximations of algebraic numbers by rationals\, which said one can't appro
ximate algebraic numbers "well" using rationals. This enabled him to const
ruct explicitly some transcendental numbers.\n\nIn this talk\, we will dis
cuss this result and an interesting way to construct transcendental number
s.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4834
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4834
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computing Equilibria in Atomic Splittable Routing Games
DTSTART;VALUE=DATE-TIME:20160208T103000Z
DTEND;VALUE=DATE-TIME:20160208T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4840@cern.ch
DESCRIPTION:An atomic splittable routing game (ASRG) is a network congesti
on problem where each player has some finite amount of flow he wants to se
nd in the network\, while minimizing his cost.\n\nIt has been known that e
quilibrium exists in ASRG games. But computing the equilibrium for ASRG ha
s been an open problem for many years.\n\nIn this report we give an algori
thm for finding the equilibrium when the underlying network is parallel ed
ge graph with linear latencies. Our algorithm is similar to an earlier alg
orithm by Huang [Theory Comp. Sys.\, 2013]\, but uses a modified support e
numeration step.\n\nNext we give an algorithm for finding the equilibrium
for two player ASRG when the underlying network is a parallel-edge graph w
ith quadratic latencies. Our algorithm is obtained by using support enumer
ation\, along with convex programming and results about the structure of e
quilibrium with quadratic latencies. We assume in this report that exact s
olutions to convex programs and binary searches can be obtained in polynom
ial time.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId
=4840
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4840
END:VEVENT
BEGIN:VEVENT
SUMMARY:Liveness-Based Garbage Collection
DTSTART;VALUE=DATE-TIME:20160209T103000Z
DTEND;VALUE=DATE-TIME:20160209T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4832@cern.ch
DESCRIPTION:Current garbage collectors leave much heap-allocated data unco
llected because they preserve data reachable from a root set. However\, on
ly live data—a subset of reachable data—need be preserved. We formula
te a context-sensitive liveness analysis for structured data and prove it
correct. We then use a 0-CFA-like conservative approximation to annotate e
ach allocation and function-call program point with a finite-state automat
on—which the garbage collector inspects to curtail reachability during m
arking. As a result\, fewer objects are marked (albeit with a more expensi
ve marker) and then preserved (e.g. by a copy phase). Experiments confirm
the expected performance benefits—increase in garbage reclaimed and a co
nsequent decrease in the number of collections\, a decrease in the memory
size required to run programs\, and reduced overall garbage collection tim
e for a majority of programs (work done with Prasanna K (IITB)\, Rahul As
ati (IITB)\, Amitabha Sanyal (IITB)\, and Alan Mycroft(University of Cambr
idge)).\n\nBio: Dr Amey Karkare completed his PhD from IIT Bombay in 2009\
, and his B.Tech. from IIT Kanpur in 1998. His areas of interest include I
ntelligent Tutoring Systems\, Program Analysis\, Compiler Optimizations\,
and Functional Programming. He has more than 7 years of industrial experie
nce most of which is in the area of Compiler Optimizations. He is an Assis
tant Professor and P K Kelkar Research Fellow in the department of CSE at
IIT Kanpur. He is currently visiting Dept of CSE\, IIT Bombay (until Dec
2016).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=48
32
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4832
END:VEVENT
BEGIN:VEVENT
SUMMARY:Language-based Security for Multi-party Applications
DTSTART;VALUE=DATE-TIME:20160222T093000Z
DTEND;VALUE=DATE-TIME:20160222T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4839@cern.ch
DESCRIPTION:Applications that combine code and data from multiple sources
are hard to program and even harder to get right. For example\, cryptograp
hic protocols for Secure Multi-party Computation (SMC) are known for a lon
g time. Yet\, existing SMC frameworks neither provide an easy-to-use progr
amming model\, nor allow for formal reasoning about the correctness and se
curity of SMC applications. As another example\, for web applications that
host content from different origins\, programmers often struggle to enfor
ce fine-grained separation between different JavaScript components\, leadi
ng to critical security vulnerabilities. In this talk\, I present my work
that addresses these problems.\n\nFirst\, I present Wys*\, a new SMC domai
n-specific language hosted in F*\, a verification-oriented programming lan
guage developed at Microsoft Research. Wys* allows programmers to program
rich\, reactive SMC applications as if they were regular\, single-threaded
F* programs\, and execute them using a low-level distributed semantics. T
he single-threaded semantics enables programmers to seamlessly use F*'s lo
gic for verifying the correctness and security properties of their SMC pro
grams. We have formally proved (again\, using F*) that the properties veri
fied for the Wys* source programs carry over when the programs are run usi
ng the low-level semantics. We have programmed and verified several SMC ap
plications in Wys*\, including a novel card dealing application.\n\nNext\,
I present Safe TypeScript\, a sound and efficient gradual type system for
TypeScript\, a typed superset of JavaScript. Safe TypeScript introduces t
wo novel ideas: differential subtyping and an erased type modality. While
differential subtyping ensures that at runtime\, objects are instrumented
with minimum amount of runtime type information needed for type safety\, t
he erased type modality provides safe and controlled type erasure along wi
th fine-grained information hiding. We have empirically evaluated Safe Typ
eScript on 120\,000 lines of existing code\, including bootstrapping the S
afe TypeScript compiler itself (90\,000 LOC)\, for which we measure a 15%
runtime overhead of type safety.\n\nBio: Aseem Rastogi is a PhD candidate
in the Computer Science Department at the University of Maryland\, College
Park. His research applies formal methods in programming languages to des
ign abstractions that simplify the task of programming software systems\,
and enable programmers to state and enforce crucial invariants such as tho
se related to the correctness and security of the software. Before joining
the University of Maryland\, Aseem obtained an MS in Computer Science fro
m the State University of New York at Stony Brook.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=4839
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4839
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rigidity and Tolerance of Point Processes
DTSTART;VALUE=DATE-TIME:20160223T103000Z
DTEND;VALUE=DATE-TIME:20160223T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4852@cern.ch
DESCRIPTION:S. Ghosh and Y. Peres showed that in certain point processes\,
the configuration outside a bounded domain determines the number of point
s of the process inside the domain (and in some cases even the center of m
ass). They named this surprising phenomenon as rigidity. In joint work wit
h Subhroshekhar Ghosh\, we systematically investigated the extent of rigid
ity and tolerance in determinantal point processes and in zeros of random
analytic functions. I shall start with the basic definitions and present s
ome of the results.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay
.py?confId=4852
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4852
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Multiple Access Server with distributed Queues
DTSTART;VALUE=DATE-TIME:20160301T103000Z
DTEND;VALUE=DATE-TIME:20160301T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4877@cern.ch
DESCRIPTION:We consider a slotted Gaussian multiple access system where i
ndependent arrivals are queued at each transmitter. The arriving packets a
t each terminal have to delivered to the receiver within a max-delay const
raint. An interesting distributed case happens when each transmitter only
knows its own arrival process. In this setup\, we seek energy efficient tr
ansmission schemes minimizing the average sum-power. Optimal schemes are e
xplicitly identified for a unit slot delay constraint\, and then extended
to general max-delay constraints using an iterative procedure.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=4877
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4877
END:VEVENT
BEGIN:VEVENT
SUMMARY:Threshold Secret Sharing Requires a Linear Size Alphabet
DTSTART;VALUE=DATE-TIME:20160302T103000Z
DTEND;VALUE=DATE-TIME:20160302T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4879@cern.ch
DESCRIPTION:A $t$-out-of-$n$ threshold secret sharing scheme is a protocol
for sharing a secret among $n$ parties so that no $t -1$ parties gain any
information about the secret but any $t$ parties can recover the secret.\
n\nWe prove that for every $n$ and $1 \n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=4879
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4879
END:VEVENT
BEGIN:VEVENT
SUMMARY:Exact and Approximation Algorithms for Weighted Matroid Intersecti
on
DTSTART;VALUE=DATE-TIME:20160315T103000Z
DTEND;VALUE=DATE-TIME:20160315T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4894@cern.ch
DESCRIPTION:We propose new exact and approximation algorithms for the weig
hted matroid intersection problem. Our exact algorithm is faster than prev
ious algorithms when the largest weight is relatively small. Our approxim
ation algorithm delivers a $(1-\\epsilon)$-approximate solution with a ru
nning time significantly faster than most known exact algorithms.\n\nThe c
ore of our algorithms is a decomposition technique: we decompose an instan
ce of the weighted matroid intersection problem into a set of instances of
the unweighted matroid intersection problem. The computational advantage
of this approach is that we can make use of fast unweighted matroid inters
ection algorithms as a black box for designing algorithms. Precisely speak
ing\, we prove that we can solve the weighted matroid intersection problem
via solving W instances of the unweighted matroid intersection problem\,
where W is the largest given weight. Furthermore\, we can find a (1-\\epsi
lon)-approximate solution via solving O(\\epsilon^{-1} \\log r) instances
of the unweighted matroid intersection problem\, where r is the smallest r
ank of the given two matroids.\n\nBio: Chien-Chung Huang is currently an a
ssistant professor at Chalmers University of Technology in Sweden. He obt
ained his Ph.D. at Dartmouth College in 2008\, under the supervision of Pe
ter Winkler. He works in the area of algorithm design nd analysis. The ma
jor topics that he has worked on include stable matching\, machine schedul
ing\, and selfish routing games.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=4894
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4894
END:VEVENT
BEGIN:VEVENT
SUMMARY:D^2 Sampling and the k-means Problem
DTSTART;VALUE=DATE-TIME:20160329T103000Z
DTEND;VALUE=DATE-TIME:20160329T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4897@cern.ch
DESCRIPTION:Given a set of points P in a d-dimensional Euclidean space\, t
he k-means clustering problem seeks to find a set C of k centers such that
the sum over all points in P of the square of the distance to the closest
center in C is minimized. A natural random sampling based heuristic for f
inding the k centers is as follows -- pick the first center uniformly at r
andom from the set of points\, the next one with probability proportional
to the square of the distance from the first center and so on. In this tal
k\, I will survey algorithms for the k-means problem based on this random
sampling technique. I will show that this idea can be used to obtain fast
PTAS for this problem. We extend this result to constrained k-means cluste
ring problems\, where there may be additional constraints on valid cluste
rings of the input points (this is joint work with Anup Bhattacharya and R
agesh Jaiswal).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=4897
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4897
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Randomized Algorithms for Online Preemptive Matching
DTSTART;VALUE=DATE-TIME:20160330T043000Z
DTEND;VALUE=DATE-TIME:20160330T053000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4932@cern.ch
DESCRIPTION:In this series of two talks\, we will investigate the power of
randomized algorithms for the maximum cardinality matching (MCM) and the
maximum weight matching (MWM) problems in the online preemptive model. In
this model\, the edges of a graph are revealed one by one and the algorith
m is required to always maintain a valid matching. On seeing an edge\, the
algorithm has to either accept or reject the edge. When an edge is accept
ed\, the edges adjacent to it are discarded. The goal is to construct as l
arge (heavy) a matching as possible.\n\nThe complexity of the problem is
settled for deterministic algorithms. For randomized algorithms\, a lower
bound of $1.693$ on the competitive ratio is known for MCM with a trivial
upper bound of two\; an upper bound of $5.356$ is known for MWM. We init
iate a systematic study of the same in this paper with an aim to isolate a
nd understand the difficulty. We begin with a primal-dual analysis of the
deterministic algorithm due to McGregor. All deterministic lower bounds a
re on instances which are trees at every step. Using a considerable exten
sion of the (simple) primal-dual analysis of the deterministic case\, we s
how a randomized algorithm that is $\\frac{28}{15}$-competitive for this c
lass of (unweighted) graphs.\n\nIn the first talk\, we will review the bac
kground and earlier results. In the second talk\, we will describe our ran
domized algorithm. We will also present other related results if time perm
its (based on joint work with Ashish Chiplunkar and Sundar Vishwanathan: h
ttp://arxiv.org/abs/1412.8615.)\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=4932
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4932
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some Lower Bounds for Depth 2 Circuits with Threshold and Mod Gate
s
DTSTART;VALUE=DATE-TIME:20160331T083000Z
DTEND;VALUE=DATE-TIME:20160331T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4936@cern.ch
DESCRIPTION:In this talk\, we will see some lower bound techniques for dep
th 2 circuits with threshold and mod gates. We focus mainly on two techniq
ues to prove lower bounds for threshold circuits: communication complexity
and random restrictions. \n\nThe communication complexity approach for pr
oving lower bounds is based on the observation that depth 2 circuits with
a threshold gate at the top and a small number of low communication gates
at the bottom have efficient randomized protocols. So\, one approach for
proving lower bounds for the above mentioned classes of circuits is to pro
duce a function which has small discrepancy. The lower bound follows from
the well known result that functions with small discrepancy have large ran
domized communication complexity. \n\nThe function with small discrepancy
is obtained by lifting a base function with large sign degree by a process
called masking. This method of lifting to obtain hard functions was used
by Krause and Pudlak to obtain functions with large sign monomial complexi
ty from a base function with large sign degree. This shows an exponential
lower bound for depth 2 threshold circuits with AND gates at the bottom. W
e will see a degree-discrepency which says that if the sign degree of the
base function is large\, then the discrepancy of it's lift is small.\n\nIn
our work\, we observed that masking the Universal Threshold function whic
h is a linear threshold function also gives rise to a function which has s
mall discrepancy. Thereby\, we show that for a lifted function to be hard\
, the base function need not have high sign degree.\n\nhttps://indico.tifr
.res.in/indico/conferenceDisplay.py?confId=4936
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4936
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Randomized Algorithms for Online Preemptive Matching
DTSTART;VALUE=DATE-TIME:20160401T103000Z
DTEND;VALUE=DATE-TIME:20160401T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4933@cern.ch
DESCRIPTION:In this series of two talks\, we will investigate the power of
randomized algorithms for the maximum cardinality matching (MCM) and the
maximum weight matching (MWM) problems in the online preemptive model. In
this model\, the edges of a graph are revealed one by one and the algorith
m is required to always maintain a valid matching. On seeing an edge\, the
algorithm has to either accept or reject the edge. When an edge is accept
ed\, the edges adjacent to it are discarded. The goal is to construct as l
arge (heavy) a matching as possible.\n\nThe complexity of the problem is
settled for deterministic algorithms. For randomized algorithms\, a lower
bound of $1.693$ on the competitive ratio is known for MCM with a trivial
upper bound of two\; an upper bound of $5.356$ is known for MWM. We init
iate a systematic study of the same in this paper with an aim to isolate a
nd understand the difficulty. We begin with a primal-dual analysis of the
deterministic algorithm due to McGregor. All deterministic lower bounds a
re on instances which are trees at every step. Using a considerable exten
sion of the (simple) primal-dual analysis of the deterministic case\, we s
how a randomized algorithm that is $\\frac{28}{15}$-competitive for this c
lass of (unweighted) graphs.\n\nIn the first talk\, we will review the bac
kground and earlier results. In the second talk\, we will describe our ran
domized algorithm. We will also present other related results if time perm
its (based on joint work with Ashish Chiplunkar and Sundar Vishwanathan: h
ttp://arxiv.org/abs/1412.8615.)\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=4933
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4933
END:VEVENT
BEGIN:VEVENT
SUMMARY:Streaming Algorithms for Embedding and Computing Edit Distance in
the Low Distance Regime
DTSTART;VALUE=DATE-TIME:20160406T103000Z
DTEND;VALUE=DATE-TIME:20160406T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4952@cern.ch
DESCRIPTION:The Hamming and the edit metrics are two common notions of mea
suring \ndistances between pairs of strings x\,y lying in the Boolean hype
rcube. The \nedit distance between x and y is defined as the minimum numbe
r of \ncharacter insertion\, deletion\, and bit flips needed for convertin
g x into \ny. Whereas\, the Hamming distance between x and y is the number
of bit \nflips needed for converting x to y. \n\nIn this talk I will pres
ent a randomized injective embedding of the edit \ndistance into the Hammi
ng distance with a small (qudratic) distortion. \nNamely\, for any satisfy
ing that their edit distance equals \, the \nHamming distance between the
embedding of and is with \nhigh probability. This improves over the distor
tion ratio of $O(\\log n \n\\log^* n)$ obtained by Jowhari (2012) for smal
l values of . Moreover\, \nthe embedding output size is linear in the inpu
t size and the embedding \ncan be computed using a single pass over the in
put. \n\nI will mention several applications for this embedding. Among our
results \nwe provide a one-pass (streaming) algorithm for edit distance r
unning in \nspace and computing edit distance exactly up-to distance . \nT
his algorithm is based on kernelization for edit distance that is of \nind
ependent (joint work with Diptarka Chakraborty and Elazar Goldenberg).\n\n
https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4952
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4952
END:VEVENT
BEGIN:VEVENT
SUMMARY:Logical Dynamics in Large Games
DTSTART;VALUE=DATE-TIME:20160412T103000Z
DTEND;VALUE=DATE-TIME:20160412T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4885@cern.ch
DESCRIPTION:How would you strategize in a game with say\, 50 players? If t
he game were repeated many times\, and you saw that your neighbour was doi
ng well in several previous rounds\, would you be tempted to imitate the n
eighbour in the next round ? Would that be rational on your part ?\n\nIn g
ames with a large number of players\, outcomes are associated not with the
actual tuple of strategies chosen by players but with the distribution of
what fraction of players choose which move. The pattern of reasoning in s
uch games is different from those in which all players know each others' t
ypes. We discuss Nash equilibria\, and some logical / automata theoretic f
ormulations of stability in such games.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=4885
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4885
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Information Processing View of Chemical Reaction Networks
DTSTART;VALUE=DATE-TIME:20160415T103000Z
DTEND;VALUE=DATE-TIME:20160415T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4980@cern.ch
DESCRIPTION:One may wonder how do micro-organisms process spatially and te
mporally extensive information about their environment and respond in a ma
nner that maximizes their fitness? Such statistical processing would have
to be carried out via biochemical reaction pathways. So a natural model of
computation to answer the above question is chemical reaction networks wi
th appropriate dynamics (given by mass action kinetics or chemical master
equation). In this model of computation we will discuss how one may:\n\n1.
(previously known result*) compute the maximum likelihood estimators for
log-linear models\n2. sample from the probability distribution of a graphi
cal model\n\n*"A Scheme for Molecular Computation of Maximum Likelihood Es
timators for Log-Linear Models" - Gopalkrishnan (http://arxiv.org/pdf/1506
.03172v1.pdf)\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=4980
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4980
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computing on a Full Memory
DTSTART;VALUE=DATE-TIME:20160420T103000Z
DTEND;VALUE=DATE-TIME:20160420T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4951@cern.ch
DESCRIPTION:Suppose that you have log(n) bits of free working memory\, plu
s an additional poly(n) bits of auxiliary memory which is *full*. Meaning\
, the auxiliary memory has some contents\, and you are allowed to read/wri
te into it\, but you must promise that by the time your computation is don
e\, the contents of the auxiliary memory have been restored to their origi
nal state.\n\nWhile it may appear at first that the full memory is useless
\, it turns out that you can make a non-trivial use of it to boost the pow
er of your computation.\n\nI will show that directed connectivity can be c
omputed in this way (with log(n) working space and poly(n) full memory)\,
and related results around this problem.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=4951
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4951
END:VEVENT
BEGIN:VEVENT
SUMMARY:Reordering Rules for Machine Translation Using Least General Gener
alizations
DTSTART;VALUE=DATE-TIME:20160422T100000Z
DTEND;VALUE=DATE-TIME:20160422T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4979@cern.ch
DESCRIPTION:Machine translation has been one of the key problems in comput
ational linguistics. An important part of the same is to get the order of
words in the translated sentence correct. Numerous approaches exist that r
ange from treating sentences as plain strings to those relying on sentence
structure. Among the latter\, manually written rules as well as rules lea
rnt by machine have been known to benefit the intended purpose.\n\nIn this
talk\, we discuss our work that learns rules for ordering words in the tr
anslated sentence using structural information about the original sentence
. We view the problem as that of learning a declarative set of instruction
s and thereby deploy a technique used in program synthesis that uses an op
erator called LGG (Least General Generalization) (this work was done joint
ly with Ganesh Ramakrishnan and Amitabha Sanyal at IIT\, Bombay).\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4979
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4979
END:VEVENT
BEGIN:VEVENT
SUMMARY:Types\, Proofs and Homotopy
DTSTART;VALUE=DATE-TIME:20160425T103000Z
DTEND;VALUE=DATE-TIME:20160425T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4988@cern.ch
DESCRIPTION:In many programming languages\, values have an associated type
that the compilers enforces. For example\, adding a value of type INTEGER
to a value of type STRING will be flagged as an error by these compilers.
Such checks by the compiler helps in eliminating a large class of bugs th
at arise in practice.\n\nA proof assistant is a program that checks a form
al proof for correctness. It turns out that verifying mathematical proofs
are closely connected to checking types in a sufficiently sophisticated pr
ogramming language. This connection goes by the name Curry-Howard isomorph
ism although several others have independently made the above connection.
One of the goals of my talk is to give an overview of this deep connection
.\n\nHaving introduced the basics of type theory\, I will look at the exci
ting new developments in this area: the theory of homotopy types and univa
lent foundation.\n\nI will try to make this talk as self contained as poss
ible. In particular no background on programming language\, type theory or
homotopy theory will be assumed.\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=4988
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4988
END:VEVENT
BEGIN:VEVENT
SUMMARY:Average Case Lower Bounds for Bounded Depth Threshold Circuits
DTSTART;VALUE=DATE-TIME:20160426T103000Z
DTEND;VALUE=DATE-TIME:20160426T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4994@cern.ch
DESCRIPTION:A threshold gate is a Boolean function that accepts its input
based on whether some weighted linear combination of its inputs exceeds so
me fixed threshold value or not. A threshold circuit is a circuit made up
of threshold gates. Proving superpolynomial lower bounds for depth-2 thres
hold circuits is one of the outstanding open problems in circuit complexit
y today.\n\nWeak lower bounds for threshold circuits have been known for t
wo decades. Impagliazzo\, Paturi\, and Saks (building on work of Paturi an
d\nSaks) showed that any threshold circuit of depth-d computing the Parity
function must have at least n^{1/(d-1)} gates and n^{1+1/exp(d)} wires. S
urprisingly these bounds are known to be nearly tight.\n\nWe show stronger
*average* case lower bounds in the same regime. We show that for each in
teger d > 1\, there is an epsilon_d > 0 such that Parity has correlation a
t most 1/n^{\\Omega(1)} with depth-d threshold circuits which have at most
n^{1/2(d-1)} gates or n^{1+epsilon_d} wires. We also show that the Genera
lized Andreev Function has correlation at most 1/2^{n^{\\Omega(1)}} with d
epth-d threshold circuits which have at most n^{1+ epsilon_d} wires.\n\nI
will talk about some of the techniques in this and other works on threshol
d circuits (joint work with Ruiwen Chen (Oxford) and Rahul Santhanam (Oxfo
rd)).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=499
4
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4994
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fence Synthesis Using Happens-before Formulas
DTSTART;VALUE=DATE-TIME:20160427T103000Z
DTEND;VALUE=DATE-TIME:20160427T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5011@cern.ch
DESCRIPTION:Weak memory adds many behaviors in a concurrent program that a
re unexpected by the most programmers. The weak memory behaviors can be re
moved by placing memory fences in the programs. For performance\, one shou
ld only add minimal number of fences such that there are no unwanted behav
iors. In this paper\, we present a novel method for synthesizing fences th
at disallow behaviors that are due to week memory and violate a safety pro
perty of a straight line concurrent program. Our method constructs a formu
la that encodes all violating executions of the program under weak memory
and iteratively constructs a happens-before formula over memory events tha
t represents memory event orderings that leads to error for some inputs. A
fterwards\, our method searches for event cycles encoded in the formula an
d optimally introduces fences such that the cycles are removed from the pr
ogram behavior. We have implemented the above method in Tara. Our tool sup
ports TSO\, PSO\, and RMO memory models. We applied Tara on benchmarks\, a
nd compared the performance with fence synthesis tools Glue and Memorax.\n
\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5011
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5011
END:VEVENT
BEGIN:VEVENT
SUMMARY:Understanding Wolfeʼs Heuristic: Submodular Function Minimization
and Projection onto Polytopes
DTSTART;VALUE=DATE-TIME:20160503T103000Z
DTEND;VALUE=DATE-TIME:20160503T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-4928@cern.ch
DESCRIPTION:Owing to several applications in large scale learning and visi
on problems\, fast submodular function minimization (SFM) has become a ver
y important problem. Theoretically\, unconstrained SFM is polynomial time
solvable\, however\, these algorithms are not practical. In 1976\, Wolfe p
roposed a heuristic for projection onto a polytope\, and in 1980\, Fujishi
ge showed how Wolfe's heuristic can be used for SFM. For general submodula
r functions\, this Fujishige-Wolfe heuristic seems to have the best empiri
cal performance. Despite its good practical performance\, very little is k
nown about Wolfe's projection algorithm theoretically.\n\nIn this talk\, I
will describe the first analysis which proves that the heuristic is polyn
omial time for bounded submodular functions. Our work involves the first c
onvergence analysis of Wolfeʼs projection heuristic\, and proving a robus
t version of Fujishigeʼs reduction theorem.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=4928
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=4928
END:VEVENT
BEGIN:VEVENT
SUMMARY:Forster's Lower Bound
DTSTART;VALUE=DATE-TIME:20160506T103000Z
DTEND;VALUE=DATE-TIME:20160506T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5025@cern.ch
DESCRIPTION:We will talk about the notion of the sign-rank of a {-1\, 1}-v
alued matrix\, which measures the robustness of it's rank under sign prese
rving changes. We will first see a neat geometric interpretation of the s
ign-rank\, and then see how showing an upper bound on the spectral norm of
A implies a lower bound on its sign-rank\, and also see implications in l
ower bounds on communication complexity and circuit complexity in certain
models.\n\nReferences:\nJurgen Forster. A Linear Lower Bound on the Unboun
ded Error Probabilistic Communication Complexity\, 2001\nSatyanarayana V.
Lokam: Complexity Bounds using Linear Algebra\n\nhttps://indico.tifr.res.i
n/indico/conferenceDisplay.py?confId=5025
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5025
END:VEVENT
BEGIN:VEVENT
SUMMARY:Parallelization of Boolean Groebner Basis Algorithm
DTSTART;VALUE=DATE-TIME:20160510T103000Z
DTEND;VALUE=DATE-TIME:20160510T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5022@cern.ch
DESCRIPTION:Groebner basis for a multivariate polynomial ideal is a finite
basis of polynomials that has many useful properties. Large memory requir
ements and computation time for most practical problems hinder the use of
Groebner basis in many areas. Distributing the memory over many cores will
take care of the large memory requirements of the algorithm. In this talk
we discuss one approach to parallel implementation of Buchberger's algori
thm for computing the basis for boolean polynomials with efficient memory
usage without considering the computation time. We propose improvements on
the data structure\, Zero Suppressed Decision Diagram\, to facilitate eff
icient storage as well as communication of polynomials among multiple core
s. The code is implemented using OpenMPI in C++ and is compared with PolyB
oRi package in Sage.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=5022
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5022
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Matchless Match
DTSTART;VALUE=DATE-TIME:20160512T120000Z
DTEND;VALUE=DATE-TIME:20160512T133000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5035@cern.ch
DESCRIPTION:There are an equal number of eligible women and men in a villa
ge. They have their preferences. The village elders wish to match them up
in such a way that no woman and man are tempted to leave the suggested mat
ch and instead join each other. In 1962\, economists and mathematicians G
ale and Shapley proposed an amazing algorithm to find a stable match. We w
ill run this algorithm and observe a striking phenomenon that was only rec
ently analysed.\n\nNo prior experience with computer science\, economics\,
mathematics or marriage will be necessary to follow the talk.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=5035
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5035
END:VEVENT
BEGIN:VEVENT
SUMMARY:Duality Theory for Optimal Transportation Problem
DTSTART;VALUE=DATE-TIME:20160513T103000Z
DTEND;VALUE=DATE-TIME:20160513T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5032@cern.ch
DESCRIPTION:In this talk we will introduce the optimal transportation prob
lem and mention some of its applications. We will then present the duality
theory for the optimal transportation problem. (We follow "Topics in Opti
mal Transportation" - Cedric Villani for this talk.)\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=5032
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5032
END:VEVENT
BEGIN:VEVENT
SUMMARY:Complexity Measures of Boolean Functions
DTSTART;VALUE=DATE-TIME:20160520T050000Z
DTEND;VALUE=DATE-TIME:20160520T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5037@cern.ch
DESCRIPTION:Boolean functions are central to computer science. This presen
tation will focus on Boolean functions from the perspective of certain mea
sures of complexity.\n\nIn the first part of the presentation\, our focus
will be two Fourier analytic complexity measures of Boolean functions: Fou
rier sparsity and Fourier dimension. We will present a near-optimal upper
bound of Fourier dimension of Boolean functions in terms of Fourier sparsi
ty.\n\nThe second part will concentrate on different query complexity meas
ures of Boolean functions: deterministic\, randomized zero-error and rando
mized bounded-error\, and relationships amongst them. We will see explicit
constructions of functions for which these measures are polynomially sepa
rated\, and more widely than was known before our work.\n\nWe present a lo
wer bound on the PTF-sparsity of the Inner product mod 2 function that is
better than the previously known bound by a factor of 5/2.\n\nFinally\, we
present an optimal hypercontractive inequality for Fourier-sparse real-va
lued functions on the Boolean hypercube.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=5037
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5037
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Sparsest Cut and Parallel Repetition
DTSTART;VALUE=DATE-TIME:20160523T083000Z
DTEND;VALUE=DATE-TIME:20160523T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5036@cern.ch
DESCRIPTION:We consider the following three problems in the areas of Algor
ithms\, Complexity theory and Streaming algorithms respectively.\n\n1) Gi
ven a graph $G=(V\,E)$\, the Sparsest Cut problem asks for a partition of
the vertices (i.e. a cut) $S \\subseteq V$ with minimum \\emph{sparsity}:
$\\phi(S) = \\frac{|E(S\, \\bar{S})|}{|S||\\bar{S}|}$. We investigate the
power of using a well-known semi-definite programming (SDP) relaxation t
o solve for the Sparsest Cut on \\emph{expander-like} graphs\, and give an
intuitive algorithm that has good guarantees for this class of graphs. Fu
rthermore\, we show that our result on expander-like graphs suitably gener
alizes a theorem due to Goemans on embedding certain metrics into $\\ell_1
$-space.\n\n2) The Parallel Repetition theorem proven by Raz is a key resu
lt that yields a PCP theorem statement useful for proving many hardness of
approximation results. Recently\, Moshkovitz gave a simple proof of a ver
sion of the Parallel Repetition theorem that suffices for many hardness r
esults\, by using a transformation called `fortification'. We investigate
this technique further and prove new results about fortification in our wo
rk. In particular\, we propose a method for fortification using spectral
expanders\, and prove its optimality. This has important consequences: it
rules out the possibility of proving stronger hardness results via fortifi
cation than the ones known earlier.\n\n3) A $n$-uniform hypergraph $H=(V\,
E)$ is said to be two-colorable if we can find a function (coloring) $\\ch
i: V \\rightarrow \\{Red\, Blue\\}$ on its vertices such that no hyperedge
is monochromatic. For $n=2$\, the problem of checking if $H$ is two-color
able is just the problem of determining if a graph is bipartite\, but even
for $n=3$\, this is NP-hard. Inspired by work in graph streaming algorith
ms\, we consider the problem of finding a two-coloring for hypergraphs (wh
ich are given to be two-colorable) in the \\emph{streaming model}. Using c
ommunication complexity techniques\, we prove strong lower bounds on the s
pace requirements of deterministic one-pass streaming algorithms that find
a two-coloring of $H$. We also provide efficient communication protocols
that utilize two-passes\, indicating that proving multi-pass lower bounds
will require new insights.\n\nhttps://indico.tifr.res.in/indico/conference
Display.py?confId=5036
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5036
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximate Modularity
DTSTART;VALUE=DATE-TIME:20160524T053000Z
DTEND;VALUE=DATE-TIME:20160524T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5041@cern.ch
DESCRIPTION:A set function on a ground set of size n is approximately modu
lar if it satisfies every modularity requirement to within an additive err
or\;approximate modularity is the set analog of approximate linearity. In
this work we study how close\, in additive error\, can approximately modul
ar functions be to truly modular functions. While approximately linear fun
ctions have been extensively studied in the literature\, there has been no
study on approximate modularity to the best of our knowledge.\n\nWe first
obtain polynomial time algorithms that\, given any approximately modular
function\, reconstructs a modular function that is $O(n^{1/2}$ close. We a
lso show an almost matching lower bound. In a striking contrast to these n
ear-tight computational reconstruction bounds\, we then show that for any
approximately modular function\, there exists a modular function that is $
O(\\log n)$-close (joint work with Flavio Chiericetti\, Abhimanyu Das\, Ra
vi Kumar in Proceedings of FOCS 2015).\n\nhttps://indico.tifr.res.in/indic
o/conferenceDisplay.py?confId=5041
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5041
END:VEVENT
BEGIN:VEVENT
SUMMARY:How to be a Great Researcher in Computer Science
DTSTART;VALUE=DATE-TIME:20160526T053000Z
DTEND;VALUE=DATE-TIME:20160526T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5028@cern.ch
DESCRIPTION:Computer science (CS) has emerged in last 40+ years to be a br
anch of science on its own. The concerns in CS involve all the way from fu
ndamentals of mathematics to designing usable social networking apps. Sinc
e CS is a relatively new science\, it has developed its own culture of doi
ng research that is quite distinct from the other fields of research. In t
his talk\, we will discuss key aspects of the culture and how to adapt one
self to be a great computer scientist.\n\nDisclaimer: This talk is from th
e view point of formal methods research.\nThe speaker has learned the less
ons presented in the talk by working with other great computer scientists.
He himself aspires to implement the lessons in his own work.\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=5028
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5028
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Random Polynomial Time Algorithm for Approximating The Volume of
Convex Bodies
DTSTART;VALUE=DATE-TIME:20160527T103000Z
DTEND;VALUE=DATE-TIME:20160527T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5044@cern.ch
DESCRIPTION:A randomized algorithm for approximating the volume of a conve
x body K in n-dimensional Euclidean space was proposed by Martin Dyer\, Al
an Frieze and Ravi Kannan in 1988. The algorithm is based on a scheme for
sampling nearly uniformly from within K. We motivate design of the algorit
hm and prove its correctness of it. The ideas used are mostly geometric an
d some basic theory of Markov Chain Monte Carlo methods.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=5044
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5044
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lazy Servers Amid Customer Arrival Uncertainty
DTSTART;VALUE=DATE-TIME:20160601T083000Z
DTEND;VALUE=DATE-TIME:20160601T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5029@cern.ch
DESCRIPTION:Consider a service utility where customers arrive over time\,
and the server is lazy and wants to serve customers with as little effort
as possible. The server\, however\, cannot deny service to any arriving cu
stomer\, and all customers have to be served that arrive before the closin
g time. Amid uncertainty of customer inter-arrival times\, the server has
to figure out the speed of service for each customer so as to minimize the
overall effort. We will discuss very simple online algorithm with concret
e guarantees on its performance even for the worst case customer inter-arr
ival times.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=5029
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5029
END:VEVENT
BEGIN:VEVENT
SUMMARY:Restricted Invertibility
DTSTART;VALUE=DATE-TIME:20160603T103000Z
DTEND;VALUE=DATE-TIME:20160603T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5056@cern.ch
DESCRIPTION:Given a matrix A which is n x m\, the image of A (viewed as a
linear operator) is the space spanned by its m columns {c_1\, ...\, c_m} .
The rank of A\, denoted r\, can be defined either as the size of the maxi
mum set of linearly independent columns in A\, or equivalently as the numb
er of non-zero singular values of A. In many cases\, the minimum non-zero
singular value of A may be very close to zero\, which makes it unsuitable
for applications.\n\nRestricted invertibility asks for the following: Is t
here a restriction of A to a large representative subset S \\subseteq [m]
of its columns\, so that the minimum non-zero singular value of A_S is far
from zero? This intuitively asks for the following: is there a large subs
et of the big columns of A that are not only linearly independent\, but al
so as 'orthogonal' to each other as possible? Spielman and Srivastava give
a beautiful proof that shows that there is such a subset S of size rough
ly equal to a quantity called the 'stable rank' of A.\n\nThis question tur
ns up in problems related to Banach space theory\, differential privacy\,
and numerical algebra systems. The proof requires only basic linear algebr
a as a prerequisite. Furthermore\, the proof technique has been used in va
rious other places: showing the existence of expanders\, and the solution
of the longstanding Kadison-Singer problem in physics. We will have a look
at the proof in this talk.\n\nhttps://indico.tifr.res.in/indico/conferenc
eDisplay.py?confId=5056
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5056
END:VEVENT
BEGIN:VEVENT
SUMMARY:Minimizing Latency in Cloud Based Systems: Replication Over Parall
el Servers
DTSTART;VALUE=DATE-TIME:20160610T103000Z
DTEND;VALUE=DATE-TIME:20160610T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5055@cern.ch
DESCRIPTION:We are in the midst of a major data revolution. The total data
generated by humans from the dawn of civilization until the turn of the n
ew millennium is now being generated every two days. Driven by a wide rang
e of data-intensive devices and applications\, this growth is expected to
continue its astonishing march\, and fuel the development of new and large
r data centers. In order to exploit the low-cost services offered by these
resource-rich data centers\, application developers are pushing computing
and storage away from the end-devices and instead deeper into the data-ce
nters. Hence\, the end-users’ experience is now dependent on the perform
ance of the algorithms used for data retrieval within the data-centers. In
particular\, providing low-latency services is critically important to th
e end-user experience for a wide variety of applications. Our goal has bee
n to develop the analytical foundations and methodologies to enable cloud
computing and storage solutions that result in low-latency services. A var
iety of cloud based systems can be modeled using multi-server\, multi queu
e queueing systems with data locality constraints. In these systems\, repl
ication (or most sophisticated coding schemes) can be used to not only imp
rove reliability but to also reduce latency. However\, delay optimality fo
r multi-server queueing systems has been a long-standing open problem\, wi
th limited results usually in asymptotic regimes. The key question is can
we design resource allocation schemes that are near optimal in distributio
n for minimizing several different classes of delay metrics that are impor
tant in web and cloud based services? In this talk\, I will overview some
of our recent research efforts at solving this problem\, provide some key
design principles\, and outline a set of what I believe are important open
problems.\n\nBio: Ness B. Shroff received his Ph.D. degree in Electrical
Engineering from Columbia University in 1994. He joined Purdue university
immediately thereafter as an Assistant Professor in the school of ECE. At
Purdue\, he became Full Professor of ECE in 2003 and director of CWSA in 2
004\, a university-wide center on wireless systems and applications. In Ju
ly 2007\, he joined The Ohio State University\, where he holds the Ohio Em
inent Scholar endowed chair in Networking and Communications\, in the depa
rtments of ECE and CSE. He holds or has held visiting chaired professor po
sitions at Tsinghua University\, Beijing\, China and Shanghai Jiaotong Uni
versity\, Shanghai\, China\, and a visiting position at the Indian Institu
te of Technology\, Bombay\, India. Dr. Shroff is currently an editor at l
arge of IEEE/ACM Trans. on Networking\, and senior editor of IEEE Transact
ions on Control of Networked Systems. He has received numerous best paper
awards for his research and listed Thomson Reuters Book on The World’s M
ost Influential Scientific Minds as well as noted as a highly cited resear
cher by Thomson Reuters (previously ISI). He also received the IEEE INFOCO
M achievement award for seminal contributions to scheduling and resource a
llocation in wireless networks.\n\nhttps://indico.tifr.res.in/indico/confe
renceDisplay.py?confId=5055
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5055
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Quest for Optimal Solutions for the Art Gallery Problem: A Pra
ctical Iterative Algorithm
DTSTART;VALUE=DATE-TIME:20160617T103000Z
DTEND;VALUE=DATE-TIME:20160617T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5065@cern.ch
DESCRIPTION:The general Art Gallery Problem (AGP) consists in finding the
minimum number of guards sufficient to ensure the visibility coverage of a
n art gallery represented by a polygon. The AGP is a well known NP-hard pr
oblem and\, for this reason\, all algorithms proposed so far to solve it a
re unable to guarantee optimality\, except in special cases. In this talk\
, I will present a new method (due to Tozoni\, de Rezende\, and de Souza)
for solving the Art Gallery Problem by iteratively generating upper and lo
wer bounds while seeking to reach an exact solution. Though there have bee
n some notable approximation algorithms for solving particular variants of
AGP (such as those by Ghosh\; Deshpande et al\; King and Kirkpatrick\; Ni
lsson et al\; Bhattacharya et al)\, this is the first exact algorithm that
has been proposed for this problem.\n\nThe idea of the new algorithm is t
o discretize both the witness set and the guard set\, and to model the res
tricted AGP as a Set Cover Problem (SCP). Firstly\, lower and upper bounds
are computed iteratively from the linear programming relaxation of the SC
P formulation. Then\, the solutions of the primal and dual linear programs
are used to guide the refinement of the witness and the guard candidate s
ets\, giving rise to larger models\, in an attempt to continuously reduce
the duality gap. Whenever convergence happens and integrality of the prima
l linear relaxation variables is obtained\, an optimal solution is found.\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5065
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5065
END:VEVENT
BEGIN:VEVENT
SUMMARY:ComputerVaadan: On Understanding and Making Indian Classical Music
Using Computers
DTSTART;VALUE=DATE-TIME:20160623T090000Z
DTEND;VALUE=DATE-TIME:20160623T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5069@cern.ch
DESCRIPTION:Musical sounds and their compositions can be analysed and synt
hesised using digital signal processing techniques. Computers are now incr
easingly used in editing\, processing and even synthesis of music. Publicl
y available programming environments enable experimentation and constructi
on of sophisticated audio processing\, visualisation and synthesis tools o
n home computers.\n\nIn this talk\, we will provide an introduction to the
models and the techniques used in digital processing and synthesis of mus
ic. We will also discuss the role of computational and machine learning te
chniques in analysing the structure and the form of musical compositions.
We will discuss the issues in applying the computer music technology to In
dian Classical Music and describe some of our experiments in this directio
n. The talk will include demonstrations of (Indian) music analysis (raag
recognition) and synthesised using some of the discussed techniques. Final
ly\, we will speculate on the question of whether computers can "understan
d" music.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=5069
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5069
END:VEVENT
BEGIN:VEVENT
SUMMARY:Hoeffding's Inequality for Sums of Weakly Dependent andom Variable
s
DTSTART;VALUE=DATE-TIME:20160624T103000Z
DTEND;VALUE=DATE-TIME:20160624T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5081@cern.ch
DESCRIPTION:Let X_1\, X_2\, ... \,X_n be\, possibly dependent\, [0\,1]-val
ued random variables. The following question is important: What is a sharp
upper bound on the probability that their sum is significantly larger (or
significantly smaller) than their mean? We see the classical result by Wa
ssily Hoeffding (1963) for independent random variables\, and generalize t
he result to several notions of weak dependence between the random variab
les.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5081
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5081
END:VEVENT
BEGIN:VEVENT
SUMMARY:New Parallel Algorithm for Bipartite Matching
DTSTART;VALUE=DATE-TIME:20160708T103000Z
DTEND;VALUE=DATE-TIME:20160708T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5104@cern.ch
DESCRIPTION:In this talk we will see an efficient parallel algorithm for t
he Bipartite Perfect Matching problem (BPM) due to Fenner\, Gurjar and Thi
erauf. The Perfect Matching problem (PM) is to decide whether a given grap
h has a perfect matching. Edmonds gave a polynomial time algorithm for the
Perfect Matching problem in 1965. However its parallel complexity is stil
l not completely resolved as of today. Fenner et al. showed that BPM can b
e solved using uniform circuits of quasi-polynomial size n^O(log n)\, and
O(log^2 n) depth where n is the number of vertices in the input graph. Pre
viously only an exponential upper bound was known on the size of such circ
uits with poly-logarithmic depth. The result of Ferner et al. is essential
ly a derandomization of the Isolation lemma based randomized algorithm due
to Mulmuley\, Vazirani and Vazirani. We will also see an elegant proof of
the Isolation Lemma due to Noam Tashma.\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=5104
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5104
END:VEVENT
BEGIN:VEVENT
SUMMARY:Subcubic Equivalences Between Matrix\, Path and Triangle Problems
DTSTART;VALUE=DATE-TIME:20160715T103000Z
DTEND;VALUE=DATE-TIME:20160715T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5114@cern.ch
DESCRIPTION:In this talk\, we will discuss subcubic equivalence between th
e following problems:\n\n- Finding the distance product of matrices\n- Dir
ected all pairs shortest paths\n- Undirected all pairs shortest paths\n- R
eplacement paths\n- Second shortest simple path\n- Detecting a triangle wi
th negative total weight\n\nAll the above mentioned problems have cubic ti
me algorithm\, but neither a cubic lower bound nor a subcubic time algorit
hm is known for any of them. The equivalence shows that either all or none
of them have a subcubic algorithm.\n\nThese equivalences appear as a part
of the 2010 FOCS paper by Virginia Williams and Ryan Williams.\n\nWe will
begin by formalizing the notion of subcubic equivalence and the distance
product of matrices.\n\nNote: These are the equivalences that were not cov
ered during the qualifiers. But we will be covering the required basics ag
ain.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5114
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5114
END:VEVENT
BEGIN:VEVENT
SUMMARY:Progression-Free Sets in $Z_p^n$\, and Applications
DTSTART;VALUE=DATE-TIME:20160722T103000Z
DTEND;VALUE=DATE-TIME:20160722T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5126@cern.ch
DESCRIPTION:In this talk\, I would like to attempt to give a brief glimpse
at and around the recent developments related to the Cap Sets problem:\n\
nRecently\, there has been a string of breakthrough results which started
with a paper of Croot-Lev-Pach\, related to bounds on progression free set
s in $Z_4^n$. This was followed up by a great improvement to a longstandin
g problem on the bound on Cap Sets (progression free sets in $Z_3^n$) by E
llenberg-Gijswijt. The proof is surprisingly simple\, and has found severa
l applications\, most notably in proving the Erdos-Szemeredi Sunflower con
jecture\, improvements in bounds on tri-coloured sum free sets\, great imp
rovements in removal lemmas in $Z_2^n$ etc..\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=5126
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5126
END:VEVENT
BEGIN:VEVENT
SUMMARY:Weierstrass Polynomial Approximation Theorem for C[0\,1] via Weak
Law of Large Numbers
DTSTART;VALUE=DATE-TIME:20160725T103000Z
DTEND;VALUE=DATE-TIME:20160725T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5112@cern.ch
DESCRIPTION:Karl Wierstrass showed that given a continuous function f on [
0\, 1] and an epsilon positive there is a polynomial p such that it is uni
formly epsilon close to f on [0\, 1]. In this talk we give a proof of thi
s using coin tossing. We then generalize this to the case of simplexes an
d hyper cubes. We also discuss approximation by C infinity functions using
Gauss kernels.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?
confId=5112
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5112
END:VEVENT
BEGIN:VEVENT
SUMMARY:Randomized Sabotage Complexity
DTSTART;VALUE=DATE-TIME:20160729T103000Z
DTEND;VALUE=DATE-TIME:20160729T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5131@cern.ch
DESCRIPTION:One of the famous open problems in randomized query complexity
is the composition question: Is R(f o g) = Omega(R(f)R(g)) for all boolea
n functions f and g. Shalev Ben-David and Robin Kothari recently showed th
at this is true for a restricted case\, where R(f o h o g) = Omega(R(f)R(h
)R(g)) for some fixed h. To do so\, they introduced randomized sabotage co
mplexity. This complexity measure has several nice properties.\n\nIn this
talk\, I will present the randomized sabotage complexity measure\, go thro
ugh some of its properties and present the proof of the restricted composi
tion theorem. Most of this will just be through making simple observations
\, almost no background is required.\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=5131
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5131
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lightweight Formal Methods for LLVM Verification
DTSTART;VALUE=DATE-TIME:20160805T093000Z
DTEND;VALUE=DATE-TIME:20160805T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5071@cern.ch
DESCRIPTION:Peephole optimizations perform local rewriting to improve the
efficiency of the code input to the compiler. These optimizations are a p
ersistent source of bugs. The talk will present Alive and Alive-FP\, doma
in-specific languages for writing integer and floating-point related peeph
ole optimizations in LLVM\, respectively. A transformation expressed in t
hese DSLs is shown to be correct automatically by encoding the transformat
ion into constraints whose validity is checked using a Satisfiability Modu
lo Theory (SMT) solver. Furthermore\, an optimization expressed in the DSL
can be automatically translated into C++ code that is suitable for inclus
ion in an LLVM optimization pass.\n\nThese DSL’s are based on an attemp
t to balance usability and formal methods. We have discovered numerous bug
s in the LLVM compiler and the Alive tool is actively used by LLVM develop
ers to check new optimizations. I will also highlight the challenges in i
ncorporating lightweight formal methods in the tool-chain of the compiler
developer and the lessons learned in this project. I will conclude by brie
fly describing other active projects in my group on approximation\, parall
el program testing\, and verification.\n\nBio: Santosh Nagarakatte is an A
ssistant Professor of Computer Science at Rutgers University. He obtained
his PhD from the University of Pennsylvania in 2012. His research interest
s are in Hardware-Software Interfaces spanning Programming Languages\, Com
pilers\, Software Engineering\, and Computer Architecture. His papers have
been selected as IEEE MICRO TOP Picks papers of computer architecture con
ferences in 2010 and 2013. He has received the NSF CAREER Award in 2015\,
ACM SIGPLAN PLDI 2015 Distinguished Paper Award\, ACM SIGSOFT ICSE 2016 Di
stinguished Paper Award\, and the Google Faculty Research Award in 2014 fo
r his research on LLVM compiler verification.\n\nhttps://indico.tifr.res.
in/indico/conferenceDisplay.py?confId=5071
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5071
END:VEVENT
BEGIN:VEVENT
SUMMARY:Bandlimited Field Estimation from Samples Recorded by Location-Una
ware Sensors
DTSTART;VALUE=DATE-TIME:20160809T103000Z
DTEND;VALUE=DATE-TIME:20160809T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5141@cern.ch
DESCRIPTION:Remote sensing with a distributed array of stationary sensors
or a mobile sensor has been of great interest. With the advent of Internet
of Things (IOTs) for sensing applications in smart cities\, the topic at
large will become more interesting. This talk will introduce and propose
solutions to a fundamental question: can a spatial field be estimated from
samples taken at unknown sampling locations?\n\nIn this talk\, we will di
scuss recent works where a spatially bandlimited field over a finite suppo
rt is sampled at unknown locations\, and the ensuing set of samples have t
o be used to estimate the spatially bandlimited field generating them. It
is assumed that the unknown sampling locations are obtained by statistica
l realization of a random process. The statistics of sampling locations i
s then leveraged to estimate the bandlimited field in question. Two models
of sampling locations will be explored: (i) a scattering scenario where s
ensors are deployed uniformly at random in an interval of interest\; and (
ii) a mobile sampling scenario where a location-unaware mobile sensor reco
rds spatial field values on a renewal process with unknown distribution.
In this unknown sampling location setup\, a _universal_ estimate for the f
ield will be developed and its mean-squared error (distortion) will be ana
lyzed as a function of the average number of field samples collected (i.e.
\, oversampling). In both these sampling models\, the effect of additive m
easurement-noise will also be examined.\n\nBio: Animesh Kumar is an Associ
ate Professor in the Department of Electrical Engineering at the Indian In
stitute of Technology Bombay. He obtained his Ph.D. degree in Electrical E
ngineering and Computer Sciences from the University of California\, Berke
ley. He is an Affiliate Member of the IEEE Signal Processing Society SPTM
Technical Committee. His current research interests include sampling theor
y and quantization\, statistical and distributed signal processing\, and T
V white space.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=5141
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5141
END:VEVENT
BEGIN:VEVENT
SUMMARY:Community Detection in Networks: Algorithms\, Complexity\, and Inf
ormation Limits
DTSTART;VALUE=DATE-TIME:20160811T060000Z
DTEND;VALUE=DATE-TIME:20160811T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5144@cern.ch
DESCRIPTION:Detecting or estimating a dense community from a network graph
offers a rich set of problems involving the interplay of algorithms\, com
plexity\, and information limits. This talk will present an overview and r
ecent results on this topic (joint work with Yihong Wu and Jiaming Xu).\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5144
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5144
END:VEVENT
BEGIN:VEVENT
SUMMARY:Higher Order Cheeger Inequalities
DTSTART;VALUE=DATE-TIME:20160812T103000Z
DTEND;VALUE=DATE-TIME:20160812T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5149@cern.ch
DESCRIPTION:Cheeger inequalities in spectral graph theory help to comment
on the approximate connectivity or expansion of a graph (a combinatorial p
roperty) from the eigenvalues of the adjacency matrix of the graph (an alg
ebraic property). More specifically\, it says that the expansion of a grap
h can be tightly bound by the second largest eigenvalue of the graph. The
Cheeger inequalities were generalized by Lee-Gharan-Trevisan so that the k
-way expansion of a graph can be approximated by the k largest eigenvalues
of the adjacency matrix.\n\nIn this talk\, i will present the higher orde
r cheeger inequality\, which roughly says\, that the first k eigenvalues o
f the laplacian of a graph are close to 0 if and only if we can find k app
roximately disjoint subsets in the graph. I will start with the basics and
show the inequality.\n\nPaper: https://arxiv.org/abs/1111.1055\n\nhttps:/
/indico.tifr.res.in/indico/conferenceDisplay.py?confId=5149
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5149
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Near Optimal Separation Between Homogeneous Depth-5 and Homogene
ous Depth-4 Arithmetic Circuits
DTSTART;VALUE=DATE-TIME:20160823T103000Z
DTEND;VALUE=DATE-TIME:20160823T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5146@cern.ch
DESCRIPTION:We will show that there is a family of n- variate polynomials
of degree d = O(log^2 n)\, which can be computed by linear sized homogeneo
us depth-5 arithmetic circuits\, where as any homogeneous depth-4 circuit
computing it must have size at least n^{Omega(sqrt d)}.\n\nThis shows that
for this range of parameters\, the upper bounds for depth reductions to h
omogeneous depth-4 circuits obtained by Agrawal-Vinay\, Koiran and Tavenas
are tight up to constants in the exponent\; even for very simple circuits
like homogeneous depth-5 arithmetic circuits. Prior to this work\, it was
known that the depth reduction results are tight up to constants in the e
xponent\, for algebraic branching programs (based on a joint work with Ram
prasad Saptharishi).\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=5146
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5146
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Polynomial Approximations to AC0
DTSTART;VALUE=DATE-TIME:20160830T110000Z
DTEND;VALUE=DATE-TIME:20160830T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5175@cern.ch
DESCRIPTION:In this talk\, we will survey questions related to polynomial
approximations of AC0. A classic result due to Tarui (1991) and Beigel\, R
eingold\, and Spielman (1991)\, that any AC0 circuit of size s and depth d
has an ε-error probabilistic polynomial over the reals of degree (log(s/
ε))^{O(d)}. We will have a re-look at this construction and show how to i
mprove the bound to (log s)^{O(d)}⋅log(1/ε)\, which is much better for
small values of ε.\n\nAs an application of this result\, we show that (lo
g s)^{O(d)}⋅log(1/ε)-wise independence fools AC0\, improving on Tal's s
trengthening of Braverman's theorem that (log(s/ε))^{O(d)}-wise independe
nce fools AC0. Time permitting\, we will also discuss some lower bounds on
the best polynomial approximations to AC0 (joint work with Srikanth Srini
vasan).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5
175
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5175
END:VEVENT
BEGIN:VEVENT
SUMMARY:Existence of Brownian Motion
DTSTART;VALUE=DATE-TIME:20160902T103000Z
DTEND;VALUE=DATE-TIME:20160902T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5180@cern.ch
DESCRIPTION:The Brownian motion is one of the most interesting and useful
of all Stochastic Processes. It has an enormous range of applications rang
ing from physics (Einstein) to finance (starting with Bachelier). In this
talk we will define and prove the existence of the Brownian motion\, by sh
owing that it can be represented as the random sum of integrals of orthogo
nal functions.\n\nThe material for this talk is taken from Michael Steele'
s textbook: Stochastic Calculus and Financial Applications (chapter 3).\n\
nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5180
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5180
END:VEVENT
BEGIN:VEVENT
SUMMARY:Topological Proof of the Compactness Theorem of Propositional Logi
c
DTSTART;VALUE=DATE-TIME:20160909T103000Z
DTEND;VALUE=DATE-TIME:20160909T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5181@cern.ch
DESCRIPTION:The compactness theorem states that there is a model for an in
finite set S of propositional formulas\, if and only if\, there is a model
for every finite subset of S. Compactness is one of the central notions o
f logic and has a wide variety of applications mainly in Model Theory.\n\n
Kurt Gödel proved the countable compactness theorem in 1930 using Mathema
tical Logic which was generalised to the uncountable case by Anatoly Malts
ev in 1936.\n\nWe will look at a topological proof of compactness\, initia
lly given for propositional logic with a set of countably infinite proposi
tional constants and then generalised to a set of propositional constants
of any size using Tychonoff Theorem. Here we give the general proof.\n\nht
tps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5181
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5181
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computing Equilibria in Atomic Splittable Routing Games
DTSTART;VALUE=DATE-TIME:20160914T083000Z
DTEND;VALUE=DATE-TIME:20160914T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5189@cern.ch
DESCRIPTION:An atomic splittable routing game (ASRG) is a network congesti
on problem where each player has some finite amount of flow he wants to se
nd in the network\, while minimizing his cost due to congestion. It is kno
wn that equilibria exist in these games\, and can be computed in polynomia
l-time for the case of linear cost functions. However\, the case of equili
brium computation with more general cost functions has been an open proble
m for many years.\n\nWe give an algorithm for computing an equilibrium in
the case of parallel-edge graphs with quadratic cost functions and two pla
yers. Our algorithm is obtained by using support enumeration\, along with
convex programming and results about the structure of equilibrium with qua
dratic cost functions. We also present some obstacles to generalizing our
algorithm to more general cases.\n\nFinally\, we present some hardness res
ults regarding equilibrium computation in more general games\, when differ
ent players may experience different delays on an edge.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=5189
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5189
END:VEVENT
BEGIN:VEVENT
SUMMARY:Complexity of Elimination
DTSTART;VALUE=DATE-TIME:20160916T103000Z
DTEND;VALUE=DATE-TIME:20160916T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5194@cern.ch
DESCRIPTION:One of the basic questions in complexity theory is how the com
plexity of computing $k$ instances of a function relates to the complexity
of computing a single instance. More precisely\, we want to know whether
we can save any computation by solving $k$ instances together instead of s
olving each instance individually. In literature jargon\, this question ha
s a name: the direct-sum problem.\n\nWe will consider an easier version of
this problem\, called the elimination problem\, in communication setting.
Consider a bipartite boolean function: $f: \\mathcal{X} \\times \\mathcal
{Y} \\rightarrow \\mathcal{R}$ where Alice gets $X \\in \\mathcal{X}$ and
Bob gets $Y \\in \\mathcal{Y}$ and jointly they compute the function $f(X\
,Y)$ by communicating with each other. In the elimination problem\, Alice
gets $X_1. \\dots\, X_k \\in \\mathcal{X}^k$ and Bob gets $Y_1\, \\dots\,
Y_k \\in \\mathcal{Y}^k$ and their goal is to output a string $\\sigma_1\,
\\dots\, \\sigma_k$ such that $f(X_i\, Y_i) \\neq \\sigma_i$ for at least
one $i \\in [k]$. Clearly this is an easier scenario than the 'direct-sum
' problem:\n\nConsider the case where $f$ is the EQUALITY function\, $X_1
= \\dots = X_k$ and $Y_1 = \\dots = Y_k$. To solve direct-sum one needs to
solve at least one instance of $f$ whereas for elimination\, no communica
tion is necessary under the said promise.\n\nWe will prove an upper and a
lower bound of the randomized communication complexity of the elimination
problem.\n\nRef: Beimel\, Amos and Daniel\, Sebastian Ben and Kushilevitz
\, Eyal and Weinreb\, Enav:\nChoosing\, Agreeing\, and Eliminating in Comm
unication Complexity.\n\nhttps://indico.tifr.res.in/indico/conferenceDispl
ay.py?confId=5194
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5194
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invariant Measures for Queueing and Spatial Markov Processes: Algo
rithms and Analysis
DTSTART;VALUE=DATE-TIME:20160921T060000Z
DTEND;VALUE=DATE-TIME:20160921T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5209@cern.ch
DESCRIPTION:As is well known\, Monte Carlo methods are ubiquitous in many
applied domains. One of the main uses of these methods is to study the lon
g-time behavior of stochastic systems of interest. Under some ergodicity c
onditions\, the long-time behavior of a system can be specified by the inv
ariant measure (or the steady-state distribution) of the Markov process th
at describes the system. In this talk\, we provide and analyze Monte Carlo
algorithms associated with invariant measures of queueing and spatial Mar
kov processes. In particular\, we propose and analyze an implementable reg
enerative simulation scheme to estimate the steady-state parameters of que
ueing networks where the inter-arrival times are generally distributed but
have exponential or heavier tails. We then consider Gibbs point processe
s (that is\, the family of spatial point processes that are absolutely con
tinuous with respect to some Poisson point process\; examples include Area
-interaction processes\, Strauss processes\, Ising models\, Loss systems).
A key feature of such a point process is that its distribution can be rea
lized as an invariant measure of a spatial birth-and-death process. We ana
lyze a simple acceptance-rejection based sampling algorithm for generating
exact samples from Gibbs point processes. We further develop and analyze
importance sampling and infinite series based representations that are see
n to provide unbiased estimators for functions of these point processes. T
he proof of their effectiveness for certain loss systems depends on our la
rge deviations analysis of no overlap probability of spheres when their ce
nters are distributed as a homogeneous Poisson point process\; this analys
is may also be of independent interest.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=5209
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5209
END:VEVENT
BEGIN:VEVENT
SUMMARY:Economics and Computation
DTSTART;VALUE=DATE-TIME:20160923T103000Z
DTEND;VALUE=DATE-TIME:20160923T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5191@cern.ch
DESCRIPTION:How should a group of friends decide which movie to watch toge
ther or which restaurant to go for dinner? How should a municipal corporat
ion decide which set of public projects to undertake? How can a organizati
onal committee allocate its funds for a set of projects that yields the ma
ximum social welfare? Can we provide a performance guarantee for these dec
isions? The research in the intersection of economics and computation has
interesting solutions for them.\n\nArtificial intelligence (AI) deals with
building systems or machines that take efficient decisions like the human
s. In a setting where multiple such decision making human/automated agents
interact\, we need to design both a robust system that provides certain p
erformance guarantees as well as help the agents to take efficient collect
ive decisions. My research considers the multi-agent systems from two comp
lementary viewpoints: (1) design protocols that are robust against any str
ategic manipulations\, and (2) use AI to assist the human agents make prov
ably efficient collective decisions. In both these settings\, individual a
gents have some private information which needs to be revealed in order to
take an efficient decision. My research in the first theme considers how
we can design mechanisms that motivates individuals to reveal their privat
e information truthfully and provide the limits of achievability of certai
n desirable properties. In the second theme\, even if the individuals had
the best intention of taking the efficient collective decision\, their lim
itations of information revelation restricts the efficiency. My research i
n this theme provides recommendation of the format of information extracti
on and gives provable guarantees to efficient decision making. In this tal
k\, I am going to provide examples and present my recent results in each o
f these two themes.\n\nBio: Swaprava is a post-doctoral fellow at the Comp
uter Science Department\, Carnegie Mellon University. Earlier\, he was a L
ecturer and Post-doctoral Fellow at the Economics and Planning Unit\, Indi
an Statistical Institute\, New Delhi. He completed his PhD from the Dept.
of Computer Science and Automation\, Indian Institute of Science\, Bangalo
re. His research interest lies in the intersection of economics and comput
ation\, applications of which are prevalent in Internet economics\, multi-
agent systems\, crowdsourcing\, resource allocation\, computational voting
\, information networks. He has been recipients of Fulbright-Nehru Post-do
ctoral Fellowship\, Tata Consultancy Services PhD Fellowship\, and Honorab
le Mention Award of Yahoo! Key Scientific Challenges Program.\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=5191
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5191
END:VEVENT
BEGIN:VEVENT
SUMMARY:Max-coloring Paths
DTSTART;VALUE=DATE-TIME:20160930T103000Z
DTEND;VALUE=DATE-TIME:20160930T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5227@cern.ch
DESCRIPTION:The max-coloring problem is to compute a legal coloring of the
vertices of a graph $G=(V\,E)$ with vertex weights $w$ such that $sum^k_{
i=1} max_{v \\in C_i} w(v_i)$ is minimized\, where $C_1\, \\ldots \,C_k$ a
re the various color classes. For general graphs\, max-coloring is as hard
as the classical vertex coloring problem. Here we study the max-coloring
problem for simple paths. Note that if the vertices are unweighted\, this
reduces to the classical vertex coloring problem and hence the solution is
to use 2 colors. However for the weighted case\, one can construct simple
instances where using more colors turns out to be optimal.\n\nIn this tal
k we will see an algorithm to optimally color the given path\, that runs i
n time $O(n + S(n))$\, where $n$ is the number of vertices and $S(n)$ is t
he time required to sort the vertex weights. We will then proceed to see a
proof of a lower bound matching this running time.\n\nThis result appears
in ISAAC 2009\, titled `Max-coloring paths: Tight bounds and extensions'
by Kavitha Telikepalli and Juli\\'an Mestre.\n\nhttps://indico.tifr.res.in
/indico/conferenceDisplay.py?confId=5227
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5227
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximating Dense Bipartite Subgraphs via an Approximate Versi
on of Caratheodory's Theorem
DTSTART;VALUE=DATE-TIME:20161007T103000Z
DTEND;VALUE=DATE-TIME:20161007T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5240@cern.ch
DESCRIPTION:We present algorithmic applications of an approximate version
of Caratheodory's theorem. The theorem states that given a set of vectors
$X$ in $\\mathbb{R^d}$\, for every vector in the convex hull of $X$ there
exists an $\\epsilon$-close (under the $p$-norm distance\, for $2 \\le p \
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5240
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5240
END:VEVENT
BEGIN:VEVENT
SUMMARY:Information Theory of DNA Shotgun Sequencing
DTSTART;VALUE=DATE-TIME:20161014T103000Z
DTEND;VALUE=DATE-TIME:20161014T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5245@cern.ch
DESCRIPTION:DNA sequencing is the basic workhorse of modern day biology an
d medicine. Shotgun sequencing is the dominant technique used: many random
ly located short fragments called reads are extracted from the DNA sequenc
e\, and these reads are assembled to reconstruct the original sequence. A
basic question is: given a sequencing technology and the statistics of the
DNA sequence\, what is the minimum number of reads required for reliable
reconstruction? This number provides a fundamental limit to the performanc
e of any assembly algorithm. In this seminar\, we will discuss a paper wit
h the above title by David Tse from 2013. For a simple statistical model o
f the DNA sequence and the read process\, We find that the answer admits a
critical phenomenon in the asymptotic limit of long DNA sequences: if the
read length is below a threshold\, reconstruction is impossible no matter
how many reads are observed\, and if the read length is above the thresho
ld\, having enough reads to cover the DNA sequence is sufficient to recons
truct. The threshold is computed in terms of the Renyi entropy rate of the
DNA sequence.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=5245
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5245
END:VEVENT
BEGIN:VEVENT
SUMMARY:Rényi Information Complexity and an Information Theoretic Charact
erization of the Partition Bound
DTSTART;VALUE=DATE-TIME:20161018T103000Z
DTEND;VALUE=DATE-TIME:20161018T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5219@cern.ch
DESCRIPTION:We introduce a new information-theoretic complexity measure IC
∞ for 2-party functions which is a lower-bound on communication complexi
ty\, and has the two leading lower-bounds on communication complexity as i
ts natural relaxations: (external) information complexity (IC) and logarit
hm of partition complexity (prt). These two lower-bounds had so far appear
ed conceptually quite different from each other\, but we show that they ar
e both obtained from IC∞ using two different\, but natural relaxations:\
n\n* The relaxation of IC∞ that yields IC is to change the order of Rén
yi mutual information used in its definition from ∞ to 1.\n\n* The relax
ation of IC∞ that yields log prt is to replace protocol transcripts used
in the definition of IC∞ with what we term "pseudotranscripts\," which
omits the interactive nature of a protocol\, but only requires that the pr
obability of any transcript given inputs x and y to the two parties\, fact
orizes into two terms which depend on x and y separately.\n\nWe also show
that if both the above relaxations are simultaneously applied to IC∞\, w
e obtain a complexity measure that is lower-bounded by the (log of) relaxe
d partition complexity\, a complexity measure introduced by Kerenidis et a
l. (FOCS 2012). We obtain a similar (but incomparable) connection between
(external) information complexity and relaxed partition complexity as Ker
enidis et al.\, using an arguably more direct proof (this is joint work wi
th Vinod Prabhakaran).\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=5219
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5219
END:VEVENT
BEGIN:VEVENT
SUMMARY:Unbounded Error Communication Complexity of XOR Functions
DTSTART;VALUE=DATE-TIME:20161021T103000Z
DTEND;VALUE=DATE-TIME:20161021T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5254@cern.ch
DESCRIPTION:We consider the unbounded error communication complexity of XO
R functions\, i.e. those of the form f of XOR\, where f is an arbitrary bo
olean function on n bits. An interesting conjecture of Zhang and Shi [ZS'0
9] asserts that for symmetric f\, the unbounded error complexity is essent
ially characterized by the number of points in {0\,...\, n-2} for which D(
i) doesn't equal D(i + 2)\, where D is the predicate corresponding to f.\n
\nWe make progress on the above conjecture by proving strong lower bounds
when f is periodic with period n^{1/2 - epsilon} for any constant epsilon
> 0. More precisely\, we show that every such XOR function has unbounded e
rror complexity n^{Omega(1)}\, unless f is a constant or parity or its com
plement\, in which case the complexity is just O(1). As a direct consequen
ce of this\, we derive new exponential lower bounds on the size of depth-2
threshold circuits computing such XOR functions (this is based on joint w
ork with Arkadev Chattopadhyay).\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=5254
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5254
END:VEVENT
BEGIN:VEVENT
SUMMARY:Impartial Peer Review
DTSTART;VALUE=DATE-TIME:20161104T103000Z
DTEND;VALUE=DATE-TIME:20161104T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5279@cern.ch
DESCRIPTION:Peer reviewing is not just an important part of research\, but
also one of the main characteristics of scientific temper. However\, over
millions of years\, we have evolved into selfish beings\, and we attempt
to maximize our benefits. This is true especially in the modern day compet
itiveness of publishing articles and grading MOOCs.\n\nUsing the naive mec
hanism\, misreporting the quality of others' work might result in personal
gain. Thus\, each reviewer has an incentive to cheat. In this talk\, we w
ill discuss a peer reviewing mechanism proposed by Kurokawa et al. which g
uarantees truthfulness. In other words\, "my reviews do not have any impac
t on my chances of being selected".\n\nhttps://indico.tifr.res.in/indico/c
onferenceDisplay.py?confId=5279
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5279
END:VEVENT
BEGIN:VEVENT
SUMMARY:Interactive Communication Over a Noisy Channel
DTSTART;VALUE=DATE-TIME:20161108T103000Z
DTEND;VALUE=DATE-TIME:20161108T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5262@cern.ch
DESCRIPTION:Alice and Bob want to hold a conversation over a noisy channel
on which adversarially chosen bits may be flipped. How can they communica
te robustly despite such an attack?\n\nWhen the conversation is one-way\,
i.e. Alice wants to send Bob a single message\, this is a well studied pro
blem\, solved by error-correcting codes. However\, these codes are not use
ful\, for instance\, when Alice and Bob need to take turns sending single
bits. To solve this problem\, Schulman (1992) invented tree codes\, showi
ng that for sufficiently small noise rates the conversation can be robustl
y simulated using a constant factor blowup in communication. Subsequently
there were a number of improvements\, culminating in a recent result of Ha
eupler (2014) conjectured to be optimal.\n\nThe drawback in the above is t
hat it depends on knowing the noise rate. We approach the problem from a d
ifferent angle\, asking what Alice and Bob can do if they have no estimate
for the noise on the channel (indeed\, if they do not even know if there
*is* any noise). We show that\, with some caveats\, even in this setting A
lice and Bob can robustly hold their conversation\, with a communication r
ate comparable to Haeupler's with respect to the actual (a posteriori) noi
se rate.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=
5262
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5262
END:VEVENT
BEGIN:VEVENT
SUMMARY:Permutations Avoiding Arithmetic Progressions
DTSTART;VALUE=DATE-TIME:20161111T103000Z
DTEND;VALUE=DATE-TIME:20161111T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5291@cern.ch
DESCRIPTION:In 1977\, J.A. Davis et al showed that any finite subset of na
tural numbers can be permuted such that it does not contain any 3-term A.P
. as a sub-sequence. However\, this is not true for the set of all natural
numbers. Furthermore\, they also show that there exists a permutation of
natural numbers which does not contain any 5-term A.P. as a sub-sequence.\
n\nGeneralizing this\, we say that a subset of natural numbers is k-avoida
ble if there exists a permutation of the elements of the set which does no
t contain any k-term A.P. as a sub-sequence. In 2008\, LeSaulnier and S. V
ijay gave bounds on the densities of subsets of natural numbers which are
3-avoidable and 4-avoidable. In this talk\, I will discuss my contribution
(in collaboration with S. Vijay) to this work\, which was an improvement
over these bounds.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.
py?confId=5291
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5291
END:VEVENT
BEGIN:VEVENT
SUMMARY:Codes\, Lower Bounds\, and Phase Transitions in the Symmetric Rend
ezvous Problem
DTSTART;VALUE=DATE-TIME:20161115T103000Z
DTEND;VALUE=DATE-TIME:20161115T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5283@cern.ch
DESCRIPTION:In the rendezvous problem\, two parties with different labelin
gs of the vertices of a complete graph are trying to meet at some vertex a
t the same time. It is well-known that if the parties have predetermined r
oles\, then the strategy where one of them waits at one vertex\, while the
other visits all n vertices in random order is optimal\, taking at most n
steps and averaging about n/2. Anderson and Weber (J. Appl. Prob. 1990\,
pp. 839–851) considered the symmetric rendezvous problem\, where both pa
rties must use the same randomized strategy. They analyzed strategies wher
e the parties repeatedly play the optimal asymmetric strategy\, determinin
g their role independently each time by a biased coin-flip. By tuning the
bias\, Anderson and Weber achieved an expected meeting time of about 0.829
n\, which they conjectured to be asymptotically optimal.\n\nWe change pers
pective slightly: instead of minimizing the expected meeting time\, we see
k to maximize the probability of meeting within a specified time T. The An
derson-Weber strategy\, which fails with constant probability when T=0(n)\
, is not asymptotically optimal for large T in this setting. Specifically\
, we exhibit a symmetric strategy that succeeds with probability 1-0(1) in
T = 4n steps. This is tight: for any a\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=5283
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5283
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Birkhoff-Von Neumann Theorem
DTSTART;VALUE=DATE-TIME:20161118T103000Z
DTEND;VALUE=DATE-TIME:20161118T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5299@cern.ch
DESCRIPTION:The Birkhoff-Von Neumann is a structure theorem characterizing
the extremal points of the convex set of doubly stochastic matrices. It p
lays a role in the theory of Doubly stochastic matrices analogous to the s
pectral theorem in the theory of Hermitian matrices. We discuss a proof of
this theorem.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=5299
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5299
END:VEVENT
BEGIN:VEVENT
SUMMARY:Randomized Rounding Revisited
DTSTART;VALUE=DATE-TIME:20161122T103000Z
DTEND;VALUE=DATE-TIME:20161122T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5228@cern.ch
DESCRIPTION:We develop new techniques for rounding packing integer program
s using iterative randomized rounding. It is based on a novel application
of multidimensional Brownian motion in $\\mathbb{R}^n$. Let $\\overset{\\s
im}{x} \\in {[0\,1]}^n$ be a fractional feasible solution of a packing con
straint $A x \\leq 1\,\\ \\ $ $A \\in {\\{0\,1 \\}}^{m\\times n}$ that max
imizes a linear objective function.\n\nOur algorithm iteratively transform
s $\\overset{\\sim}{x}$ to $\\hat{x} \\in {\\{ 0\,1\\}}^{n}$ using a rando
m walk\, such that the expected values of $\\hat{x}_i$'s are consistent wi
th the Raghavan-Thompson rounding. In addition\, it gives us intermediate
values $x'$ which can then be used to bias the rounding towards a superior
solution. Our algorithm gradually sparsifies $A$ to $A' \\in {\\{0\,1 \
\}}^{m\\times n}$ where each row in $A'$ has $\\leq \\log n$ non-zero coef
ficients with $A'\\cdot x' \\leq O(1)$. The reduced dependencies between t
he constraints of the sparser system can be exploited using {\\it Lovasz L
ocal Lemma}. Using the Moser-Tardos' constructive version\, $x'$ converges
to $\\hat{x}$ in polynomial time to a distribution over the unit hypercub
e ${\\cal H}_n = {\\{0\,1 \\}}^n$ such that the expected value of any line
ar objective function over ${\\cal H}_n$ equals the value at $\\overset{\\
sim}{x}$.\n\nWe discuss application of these techniques when $A$ is a rand
om matrix and also for a more general situation of a $k$-column sparse ma
trix (joint work with Dhiraj Madan).\n\nhttps://indico.tifr.res.in/indico/
conferenceDisplay.py?confId=5228
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5228
END:VEVENT
BEGIN:VEVENT
SUMMARY:Impact of Community Structure on Cascades
DTSTART;VALUE=DATE-TIME:20161124T083000Z
DTEND;VALUE=DATE-TIME:20161124T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5233@cern.ch
DESCRIPTION:The threshold model is widely used to study the propagation of
opinions and technologies in social networks. In this model individuals a
dopt the new behavior based on how many neighbors have already chosen it.
We study cascades under the threshold model on sparse random graphs with c
ommunity structure to see whether the existence of communities affects the
number of individuals who finally adopt the new behavior. Specifically\,
we consider the permanent adoption model where nodes that have adopted the
new behavior cannot change their state. When seeding a small number of ag
ents with the new behavior\, the community structure has little effect on
the final proportion of people that adopt it\, i.e.\, the contagion thresh
old is the same as if there were just one community. On the other hand\, s
eeding a fraction of population with the new behavior has a significant im
pact on the cascade with the optimal seeding strategy depending on how str
ongly the communities are connected. In particular\, when the communities
are strongly connected\, seeding in one community outperforms the symmetri
c seeding strategy that seeds equally in all communities.\n\nThis talk we
will be highlighting the use of the differential equation method\, also kn
own as the mean-field method\, and the use of dynamical systems theory for
the analysis and control of such complex systems (this is joint work with
Mehrdad Moharrami and Mingyan Liu at the University of Michigan and Marc
Lelarge at ENS and INRIA Paris).\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=5233
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5233
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Isoperimetric Profiles and Computational Complexity
DTSTART;VALUE=DATE-TIME:20161125T103000Z
DTEND;VALUE=DATE-TIME:20161125T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5304@cern.ch
DESCRIPTION:The isoperiemtric profile of a Graph is a function that measur
es\, for an integer k\, is the size of the smallest edge boundary over all
sets of vertices of size k.\n\nArithmetic circuits and Branching programs
are two computational models for computing polynomials. Branching program
s can be efficiently simulated by Circuits. But\, we don't know if Circuit
s are strictly more powerful than branching programs.\n\nIn this talk\, we
will see the usage of isoperimetric profiles to prove a sharp super-polyn
omial separation between monotone arithmetic circuits and monotone arithme
tic branching programs. This is a result of Hrubes and Yehudayoff [ICALP 2
016].\n\nA key ingredient in the proof is an accurate analysis of isoperim
etric profile of finite full binary trees.\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=5304
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5304
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dual Capacity Upper Bounds for Runlength Constrained Channels
DTSTART;VALUE=DATE-TIME:20161129T103000Z
DTEND;VALUE=DATE-TIME:20161129T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5229@cern.ch
DESCRIPTION:Characterising the capacity of channels with memory is a chall
enging and interesting problem in information theory. One example is a cha
nnel with an input runlength constraint. The runlength constraint is popul
ar in data storage applications and limits the minimum and maximum number
of 0s between consecutive 1s in the input sequence. The noiseless capacity
or enumeration of runlength constrained sequences was presented by Shanno
n in his 1948 paper. However\, for a noisy channel with an input runlength
constraint\, the capacity has been difficult to characterise with attempt
s by several researchers over the last 35 years or so. In this talk\, we w
ill describe the dual method for obtaining upper bounds on capacity and di
scuss its application to noisy runlength-constrained channels. We will sho
w that the dual upper bound for noisy runlength-constrained channels reduc
es to simple (and sometimes elegant) algebraic expressions for discrete ch
annels\, and is very close to achievable rates for several channels of the
oretical and practical importance. We will conclude by briefly discussing
other applications of the dual bound including peak-power limited Gaussian
channels and ISI channels.\n\nBio: Andrew Thangaraj received his B.Tech i
n Electrical Engineering from the Indian Institute of Technology (IIT)\, M
adras\, India in 1998 and a PhD in Electrical Engineering from the Georgia
Institute of Technology\, Atlanta\, USA in 2003. He was a post-doctoral r
esearcher at the GTL-CNRS Telecom lab at Georgia Tech Lorraine\, Metz\, Fr
ance from August 2003 to May 2004. From June 2004\, he has been with the D
epartment of Electrical Engineering\, IIT Madras\, where he is currently a
professor. Since Jan 2012\, he has been serving as Editor for the IEEE Tr
ansactions on Communications. His research interests are in coding theory\
, information theory and information-theoretic aspects of cryptography.\n\
nSince Oct 2011\, he has been serving as NPTEL coordinator at IIT Madras.
He has played a key role in initiating and running NPTEL online courses an
d certification. He is currently a National MOOCs Coordinator for NPTEL in
the SWAYAM project of the MHRD.\n\nhttps://indico.tifr.res.in/indico/conf
erenceDisplay.py?confId=5229
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5229
END:VEVENT
BEGIN:VEVENT
SUMMARY:Elimination and Weak Regularity
DTSTART;VALUE=DATE-TIME:20161202T103000Z
DTEND;VALUE=DATE-TIME:20161202T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5331@cern.ch
DESCRIPTION:We will consider the problem of elimination in communication c
omplexity\, that was first raised by Ambainis et al. and later studied by
Beimel et al. for its connection to the famous direct sum question. In thi
s problem\, let $f : \\{0\, 1\\}^{2n} \\rightarrow \\{0\, 1\\}$ be any boo
lean function. Alice and Bob get k inputs $x_1 \, \\cdots\, x_k$ and $y_1
\, \\cdots\, y_k$ respectively\, with $x_i \, y_i \\in \\{0\, 1\\}^n$ . Th
ey want to output a $k$-bit vector $v$\, such that there exists one index
$i$ for which $v_i = f (x_i \, y_i )$. We will prove a general result lowe
r bounding the randomized communication complexity of the elimination prob
lem for $f$ using its discrepancy. Consequently\, we will obtain strong lo
wer bounds for the functions Inner-Product and Greater-Than. To prove this
result\, we will use a pseudo-random notion called regularity that was fi
rst used by Raz and Wigderson. We will show that functions with small disc
repancy are regular. We will also observe that a weaker notion\, that we c
all weak-regularity\, already implies hardness of elimination.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=5331
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5331
END:VEVENT
BEGIN:VEVENT
SUMMARY:Strong Fooling Sets for Multi-Player Communication with Applicatio
ns to Deterministic Estimation of Stream Statistics
DTSTART;VALUE=DATE-TIME:20161206T103000Z
DTEND;VALUE=DATE-TIME:20161206T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5322@cern.ch
DESCRIPTION:We develop a paradigm for studying multi-player deterministic
communication\, based on a novel combinatorial concept that we call a {\\e
m strong fooling set}. Our paradigm leads to optimal lower bounds on the p
er-player communication required for solving multi-player $\\textsc{equali
ty}$ problems in a private-message setting. This in turn gives a very stro
ng---$O(1)$ versus $\\Omega(n)$---separation between private-message and o
ne-way blackboard communication complexities. Applying our communication c
omplexity results\, we show that for deterministic data streaming algorith
ms\, even loose estimations of some basic statistics of an input stream re
quire large amounts of space. For instance\, approximating the frequency m
oment $F_k$ within a factor $\\alpha$ requires $\\Omega(n/\\alpha^{1/(1-k)
})$ space for $k < 1$ and roughly $\\Omega(n/\\alpha^{k/(k-1)})$ space for
$k > 1$. In particular\, approximation within any {\\em constant} factor
$\\alpha$\, however large\, requires {\\em linear space\, with the trivial
exception of $k = 1$. This is in sharp contrast to the situation for rand
omized streaming algorithms\, which can approximate $F_k$ to within $(1\\p
m\\varepsilon)$ factors using $\\widetilde{O}(1)$ space for $k \\le 2$ and
$o(n)$ space for all finite $k$ and all constant $\\varepsilon > 0$. Prev
ious linear-space lower bounds for deterministic estimation were limited t
o small factors $\\alpha$\, such as $\\alpha < 2$ for approximating $F_0$
or $F_2$. We also provide certain space/approximation tradeoffs in a deter
ministic setting for the problems of estimating the empirical entropy of a
stream as well as the size of the maximum matching and the edge connectiv
ity of a streamed graph (this is joint work with Amit Chakrabarti).\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5322
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5322
END:VEVENT
BEGIN:VEVENT
SUMMARY:Efficient Redundancy Techniques to Reduce Delay in Cloud Systems
DTSTART;VALUE=DATE-TIME:20161219T083000Z
DTEND;VALUE=DATE-TIME:20161219T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5344@cern.ch
DESCRIPTION:Ensuring fast and seamless service to users is critical for to
day's cloud services. However\, guaranteeing fast response can be challeng
ing due to random service delays that are common in today's data centers.
In this talk I explore the use the redundancy to combat such service varia
bility. For example\, replicating a computing task at multiple servers and
then waiting for the earliest copy saves service time. But the redundant
tasks can cost more computing resources and also delay subsequent tasks. I
present a queueing-theoretic framework to answer fundamental questions su
ch as:\n\n1) How many replicas to launch?\n2) Which queues to join?\n3) Wh
en to issue and cancel the replicas?\n\nThis framework reveals surprising
regimes where replication reduces both delay as well as resource cost. The
task replication idea can also be generalized to analyze latency in conte
nt download from erasure coded storage. More broadly\, this work lays the
theoretical foundation for studying queues with redundancy\, uncovering ma
ny interesting future directions in cloud infrastructure\, crowdsourcing a
nd beyond.\n\nBio: Gauri Joshi is a Research Staff Member at IBM T. J. Wat
son Research Center. She will be joining the Carnegie Mellon ECE departmen
t as assistant professor in Fall 2017. Gauri completed a Ph.D from MIT EEC
S in June 2016. Before coming to MIT\, she completed a B.Tech and M. Tech
in Electrical Engineering from the Indian Institute of Technology (IIT) Bo
mbay in 2010. Her awards and honors include the Best Thesis Prize in Compu
ter science at MIT (2012)\, Institute Gold Medal of IIT Bombay (2010)\, Cl
aude Shannon Research Assistantship (2015-16)\, and Schlumberger Faculty f
or the Future fellowship (2011-2015).\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=5344
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5344
END:VEVENT
BEGIN:VEVENT
SUMMARY:Universality of Power-of-$d$ Load Balancing in Many-Server Systems
DTSTART;VALUE=DATE-TIME:20161221T103000Z
DTEND;VALUE=DATE-TIME:20161221T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5310@cern.ch
DESCRIPTION:We consider a system of $N$~parallel single-server queues with
unit exponential service rates and a single dispatcher where tasks arrive
as a Poisson process of rate $\\lambda(N)$. When a task arrives\, the dis
patcher assigns it to a server with the shortest queue among $d(N)$ random
ly selected servers ($1 \\leq d(N) \\leq N$). This load balancing strategy
is referred to as a JSQ($d(N)$) scheme\, marking that it subsumes the cel
ebrated Join-the-Shortest Queue (JSQ) policy as a crucial special case for
$d(N) = N$. We construct a stochastic coupling to bound the difference in
the queue length processes between the JSQ policy and a scheme with an ar
bitrary value of $d(N)$. We use the coupling to derive the fluid limit in
the regime where $\\lambda(N) / N \\to \\lambda < 1$ as $N \\to \\infty$ w
ith $d(N) \\to\\infty$\, along with the associated fixed point. The fluid
limit turns out not to depend on the exact growth rate of $d(N)$\, and in
particular coincides with that for the ordinary JSQ policy. We further lev
erage the coupling to establish that the diffusion limit in the critical r
egime where $(N - \\lambda(N)) / \\sqrt{N} \\to \\beta > 0$ as $N \\to \\i
nfty$ with $d(N)/(\\sqrt{N} \\log (N))\\to\\infty$ corresponds to that for
the JSQ policy. These results indicate that the optimality of the JSQ pol
icy can be preserved at the fluid-level and diffusion-level while reducing
the overhead by nearly a factor O($N$) and O($\\sqrt{N}/\\log(N)$)\, resp
ectively.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId
=5310
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5310
END:VEVENT
BEGIN:VEVENT
SUMMARY:Secure Computation of Two-Party Asymmetric Randomized Functions
DTSTART;VALUE=DATE-TIME:20161223T083000Z
DTEND;VALUE=DATE-TIME:20161223T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5358@cern.ch
DESCRIPTION:Characterization of securely computable two-party functions is
one of the most fundamental problem in cryptography. For deterministic fu
nctions\, a characterisation has been known since late 80's\; but finding
such a characterisation for {\\em randomised} functions in general remains
elusive since then. In this work we consider a special case\, where only
one party computes the output. The problem is specified by a pair $(p_{XY}
\,p_{Z|XY})$\, where $p_{XY}$ is the input distribution from which inputs
$x$ and $y$ of Alice and Bob\, respectively\, are drawn from\, and $p_{Z|X
Y}$ is the distribution according to which Bob produces his output $z$. An
y protocol for securely computing $(p_{XY}\,p_{Z|XY})$ must satisfy two pr
operties: (i) computation should be correct\, and (ii) at the end of the p
rotocol\, Alice does not learn anything about Bob's input and output\, and
Bob does not learn anything about Alice's input.\n\nWe consider two scena
rios: (i) Perfectly secure computation -- where Alice and Bob get a single
pair of inputs and Bob produces the output securely\, and (ii) Asymptotic
ally secure computation -- where Alice and Bob get blocks of inputs $(x_1\
,x_2\,\\hdots\,x_n)$ and $(y_1\,y_2\,\\hdots\,y_n)$\, respectively\, and B
ob produces $(\\hat{z}_1\,\\hat{z}_2\,\\hdots\,\\hat{z}_n)$ as his output.
Asymptotic correctness means that the $L_1$-distance between the ideal ou
tput distribution and the actual output distribution goes to zero as block
-length $n$ tends to infinity\; and asymptotic privacy means that the info
rmation leakage goes to zero as $n$ tends to infinity. In perfect security
setting we give a characterization of securely computable {\\em randomize
d} functions and prove matching lower and upper bounds on communication co
mplexity for such functions. We prove that the same characterization holds
in asymptotic security setting as well and prove matching lower and upper
bounds on communication complexity for such functions.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=5358
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5358
END:VEVENT
BEGIN:VEVENT
SUMMARY:Relating Communication Protocols and Polynomials
DTSTART;VALUE=DATE-TIME:20161230T103000Z
DTEND;VALUE=DATE-TIME:20161230T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5375@cern.ch
DESCRIPTION:Lee and Zhang showed that the communication complexity of "f c
omposed with g" is high when f is hard to approximate with a low degree po
lynomial (also g has to be from a good class of functions\, more details w
ill be given in the talk). In this talk\, we will look at a proof of this
result by transforming a communication protocol for "f composed with g" in
to a polynomial for f.\n\nhttps://indico.tifr.res.in/indico/conferenceDisp
lay.py?confId=5375
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5375
END:VEVENT
BEGIN:VEVENT
SUMMARY:Linear Sketching and One-way Communication
DTSTART;VALUE=DATE-TIME:20170106T103000Z
DTEND;VALUE=DATE-TIME:20170106T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5394@cern.ch
DESCRIPTION:Linear sketch complexity of a Boolean function f on a set of n
inputs\, introduced by Kannan\, Mossel and Yaroslavtsev\, is\, in informa
l terms\, the smallest integer d such that the value of the function can b
e concluded with high probability from the evaluation of d random F_2-line
ar functions (or parities) of the input. Linear sketch complexity of a Boo
lean function f can be easily seen to upper bound the one-way communicatio
n complexity of the function F(x\,y):=f(x+y) ('+' denotes the bit-wise xor
of two strings here). This is because given a cheap linear sketching\, th
e players of the communication game can easily simulate it with communicat
ion comparable to the cost of the sketch. The authors of the above-mention
ed work conjectured that linear sketch complexity is also a lower-bound on
the complexity of this communication problem. In other words\, linear ske
tch complexity characterizes the complexity of the communication problem.
The motivation of this conjecture comes from the observation that all know
n communication protocols are obtained from some underlying linear sketchi
ng. The authors proved a similar statement when the inputs are sampled fro
m uniform distribution. We improve on the parameters of their result in th
is work. Our proof uses Fourier analysis of Boolean functions and informat
ion theory.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=5394
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5394
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Fine-Grained Approach for Designing (Time and Space) Efficient A
lgorithms
DTSTART;VALUE=DATE-TIME:20170110T103000Z
DTEND;VALUE=DATE-TIME:20170110T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5376@cern.ch
DESCRIPTION:The classical approach for designing algorithms measures the t
ime complexity only as a measure of the input size. In the 90's\, Downey a
nd Fellows proposed a fine-grained approach for NP-hard problems: one or m
ore parameters of the input instance are defined\, and we investigate the
effect of these parameters on the running time. The goal is to design algo
rithms that work efficiently if the parameters of the input instance are s
mall\, even if the size of the input is large. Formally\, a problem is sai
d to be fixed-parameter tractable (FPT) with respect to parameter k if the
problem can be solved in time f(k)·n^{O(1)} where f is a computable func
tion and n is the input size.\n\nIn the first part of the talk\, I will gi
ve a brief overview of FPT algorithms. This active area of research has se
en many new developments over the last few years. The standard techniques
for designing FPT algorithms on undirected graphs such as bidimensionality
\, treewidth-based dynamic programming\, etc. seem to break down for direc
ted graphs. I will present some of my work which overcomes this barrier\,
and designs optimal FPT and XP algorithms for cut and connectivity problem
s on directed graphs.\n\nThe second part of the talk will be about some re
cent work on parameterized streaming algorithms. For the Maximum Matching
problem\, I will describe optimal streaming algorithms in various streamin
g models such as insertion-only\, promised and general insertion-deletion.
\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5376
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5376
END:VEVENT
BEGIN:VEVENT
SUMMARY:Constrained Counting and Sampling: Bridging the Gap between Theory
and Practice
DTSTART;VALUE=DATE-TIME:20170111T053000Z
DTEND;VALUE=DATE-TIME:20170111T070000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5381@cern.ch
DESCRIPTION:Constrained counting and sampling are two fundamental problems
in Computer Science with numerous applications\, including network reliab
ility\, decision making under certainty\, probabilistic reasoning\, and co
nstrained-random verification. In constrained counting\, the task is to co
mpute the total weight\, subject to a given weighting function\, of the se
t of solutions of the given constraints. In constrained sampling\, the tas
k is to sample randomly\, subject to a given weighting function\, from the
set of solutions to a set of given constraints.\n\nIn this talk\, I will
introduce a novel algorithmic framework for constrained sampling and count
ing that combines the classical algorithmic technique of universal hashing
with the dramatic progress made in Boolean reasoning over the past two de
cades. This has allowed us to obtain breakthrough results in constrained
sampling and counting\, providing a new algorithmic toolbox in design veri
fication\, machine learning\, probabilistic reasoning\, and the like. I w
ill demonstrate the utility of the above techniques on various real applic
ations including probabilistic inference\, hardware verification\, and our
ongoing collaboration in estimating the reliability of critical infrastru
cture networks during natural disasters.\n\nBio: Kuldeep Meel is a final y
ear Ph.D. candidate at Rice University working with Prof. Moshe Vardi and
Prof. Supratik Chakraborty (IITB). He obtained a B.Tech. from IIT Bombay a
nd an M.S. from Rice in 2012 and 2014 respectively. His research broadly l
ies at the intersection of artificial intelligence and formal methods. He
is the recipient of a 2016-17 IBM Ph.D. Fellowship\, the 2016-17 Lodieska
Stockbridge Vaughn Fellowship\, and the 2013-14 Andrew Ladd Fellowship. Hi
s research won the best student paper award at the International Conferenc
e on Constraint Programming 2015. He co-won the 2014 Vienna Center of Log
ic and Algorithms International Outstanding Masters Thesis Award.\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5381
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5381
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Perfect Graph Theorem
DTSTART;VALUE=DATE-TIME:20170113T103000Z
DTEND;VALUE=DATE-TIME:20170113T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5410@cern.ch
DESCRIPTION:A graph is said to be perfect if the chromatic number of every
induced subgraph equals its clique number. An induced cycle of size 4 or
more is called a hole and an induced subgraph that is the complement of a
cycle of size 4 or more is called an antihole. A Berge graph is a graph wi
thout any odd hole or odd antihole.\n\nThe strong perfect graph theorem st
ates that a graph is perfect if and only if it is Berge. It was conjecture
d by Claude Berge in 1961\, and proved by Maria Chudnovsky\, Neil Robertso
n\, Paul Seymour and Robin Thomas. The four authors presented their work a
t a workshop held from Oct 30 to Nov 3\, 2002 at the American Institute of
Mathematics and published a 178-page paper in 2006.\n\nFor the purpose of
this talk\, however\, we will be slightly modest and discuss a proof of t
he weak perfect graph theorem\, which states that if a graph is perfect\,
then so is its complement. Also conjectured by Berge\, it was proved by Lo
vasz in 1972.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?co
nfId=5410
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5410
END:VEVENT
BEGIN:VEVENT
SUMMARY:Throughput-Optimal Algorithms for Generalized Flow Problems
DTSTART;VALUE=DATE-TIME:20170120T060000Z
DTEND;VALUE=DATE-TIME:20170120T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5404@cern.ch
DESCRIPTION:One of the fundamental problems in Computer Networking is to e
fficiently transport packets belonging to different sessions\, such as uni
cast\, broadcast\, multicast and anycast\, collectively known as the gener
alized flows. The problem is to design an efficient routing\, and wireless
link scheduling policy\, which achieves the entire capacity region. Curre
ntly\, a throughput-optimal algorithm (Backpressure) is known only for the
unicast problem and little progress has been made towards a universal alg
orithm addressing the generalized flow problem. In this thesis\, we propos
e provably optimal algorithms for the broadcast and the generalized flow p
roblems.\n\nWe begin our study with the problem of optimal broadcasting in
a wireless Directed Acyclic Graph (DAG). Existing policies achieve the br
oadcast capacity by balancing traffic over a set of spanning trees\, which
are difficult to maintain in a large and time-varying network. We propose
a fundamentally new broadcast policy\, which is decentralized\, utilizes
local queue-length information only\, does not require the use of global t
opological structures\, such as spanning trees and offers the useful featu
re of in-order delivery of packets. It also yields a new characterization
of broadcast capacity in wireless DAGs\, which further leads to an efficie
nt algorithm for computing the capacity under the primary interference con
straints. The algorithm extends gracefully to time-varying wireless networ
ks. We next study the problem of broadcasting in networks with arbitrary t
opology and derive a new dynamic broadcast policy which can be viewed as
“Backpressure on sets”. This yields an efficient solution to the probl
em when combined with a multi-class in-order packet scheduling rule.\n\nFi
nally\, we study the generalized flow problem and derive an online dynamic
policy\, called Universal Max-Weight (UMW)\, which yields an efficient so
lution. To the best of our knowledge\, UMW is the first throughput-optimal
algorithm of such versatility in this context. Conceptually\, the UMW pol
icy is derived by relaxing the precedence constraints associated with mult
i-hop routing\, and then solving a min-cost routing and max-weight schedul
ing problem on a virtual network of queues. When specialized to the unicas
t setting\, the UMW policy yields a throughput-optimal cycle-free routing
and link scheduling policy. This is in contrast with the Backpressure poli
cy which allows for packet cycling\, resulting in excessive delay. The pro
posed algorithmic paradigm is surprisingly general and yields solutions to
other related problems\, such as optimal broadcasting with point-to-multi
point links. The proof of throughput-optimality of the UMW policy combines
techniques from stochastic Lyapunov theory with a sample path argument fr
om adversarial queueing theory and may be of independent theoretical inter
est.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5404
LOCATION: D-405 (D-Block Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5404
END:VEVENT
BEGIN:VEVENT
SUMMARY:Galvin's Proof of Dinitz's Conjecture
DTSTART;VALUE=DATE-TIME:20170120T103000Z
DTEND;VALUE=DATE-TIME:20170120T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5441@cern.ch
DESCRIPTION:An n × n Latin square is a grid with n rows and n columns suc
h that each cell of the grid is filled with one number from the set {1\,2\
,...n} and no number is repeated in any row or any column. For instance\,
any solved Sudoku puzzle is a 9 × 9 Latin square. In 1979\, Jeff Dinitz p
roposed the existence of "generalized Latin squares"\, which became known
as the Dinitz conjecture.\n\nThe Dinitiz conjecture states that given an n
× n grid\, a number m ≥ n\, and for each cell of the grid an n-element
subset of {1\,2\,...\,m}\, it is possible to fill each cell with one of t
hose elements such that no number is repeated in any row or any column. In
1994\, Fred Galvin proved Dinitz's conjecture. Doron Zeilberger wrote of
Galvin's proof\, "When I finished reading and digesting the proof\, I kick
ed myself. I felt that I could have found it myself ... With the very gene
rous help of Lady Hindsight\, I will now describe how I (and you!) could h
ave\, and should have\, found the very same proof ...".\n\nAnd that's exac
tly what we will do in this talk!\n\nhttps://indico.tifr.res.in/indico/con
ferenceDisplay.py?confId=5441
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5441
END:VEVENT
BEGIN:VEVENT
SUMMARY:Progress in Error-Correction: New Codes for Old Noise Models
DTSTART;VALUE=DATE-TIME:20170124T103000Z
DTEND;VALUE=DATE-TIME:20170124T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5380@cern.ch
DESCRIPTION:Error-correcting codes play a crucial role in safeguarding dat
a against the adverse effects of noise during communication and storage. T
hey are also powerful tools that underlie several advances in theoretical
computer science. The central challenge in coding theory is to construct c
odes with minimum possible redundancy for different noise models and requi
rements on the decoder\, along with efficient algorithms for error-correct
ion using those codes. Much progress has been made toward this quest in th
e nearly 70 years since the birth of coding theory. Several fundamental pr
oblems\, however\, continue to challenge us\, and exciting new questions r
outinely emerge to address the demands of modern technologies and applicat
ions in complexity theory/cryptography. This talk will survey some of our
recent works on error-correction in various noise models\, such as:\n\n- w
orst-case errors\, where we construct list decodable codes with redundancy
as small as the target error fraction\;\n\n- i.i.d. errors\, where we sho
w polar codes enable efficient error-correction even as the redundancy app
roaches Shannon capacity\;\n\n- bit deletions\, where we give codes that c
an correct the largest known fraction of deletions\;\n\n- single symbol er
asure\, a model of substantial current interest for tackling node failures
in distributed storage\, where we give novel repair algorithms for Reed-S
olomon codes as well as simple new codes with low-bandwidth repair mechani
sms.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5380
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5380
END:VEVENT
BEGIN:VEVENT
SUMMARY:Subspace Designs
DTSTART;VALUE=DATE-TIME:20170125T113000Z
DTEND;VALUE=DATE-TIME:20170125T133000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5448@cern.ch
DESCRIPTION:A subspace design is a (large) collection of subspaces of F_q^
n\, each of small co-dimension\, such that for any low-dimensional subspac
e W\, only a small number from the collection have a non-trivial intersect
ion with W. This notion was put forth in the context of algebraic list dec
oding where it enabled the construction of optimal redundancy list-decodab
le codes over small alphabets and for the rank-metric. An explicit constru
ction of subspace designs over large fields with near-optimal parameters w
as later given based on polynomials\; curiously\, this construction is clo
sely related to folded Reed-Solomon codes\, the list-decodable codes which
motivated subspace designs in the first place.\n\nSubspace designs have s
ince been found to play a central role in the emerging theory of "linear-a
lgebraic pseudorandomness” which aims to understand analogs of Boolean p
seudorandom objects where the rank of subspaces plays the role of the size
of subsets. In particular\, they yield “rank condensers” which in tur
n enable construction of dimension expanders\, the linear-algebraic analog
of (vertex) expanders. Via this connection\, the known explicit construct
ion of subspace designs yield simple constant-degree dimension expanders o
ver F_q^n when q \\ge poly(n). Subspace designs over any field were recent
ly constructed based on algebraic function fields\; while this incurs a sl
ight worsening of parameters\, it raises hope that perhaps one can get con
stant-degree dimension expanders over any field via this route (the only k
nown explicit construction for arbitrary fields\, via monotone expanders\,
is quite complicated).\n\nThis talk will survey these developments revolv
ing around subspace designs\, their construction\, and connections (based
on several joint works with Chaoping Xing\, Swastik Kopparty\, Michael For
bes\, and Chen Yuan).\n\nhttps://indico.tifr.res.in/indico/conferenceDispl
ay.py?confId=5448
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5448
END:VEVENT
BEGIN:VEVENT
SUMMARY:An Invitation to Causal Inference
DTSTART;VALUE=DATE-TIME:20170127T103000Z
DTEND;VALUE=DATE-TIME:20170127T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5452@cern.ch
DESCRIPTION:Consider a system consisting of a binary "stimulus" variable X
\, (e.g. whether an individual consumes tobacco)\, a binary "response" var
iable Y\, (e.g. whether the individual develops a particular disease) and
a "hidden" confounding variable U\, (e.g. a hypothetical genetic factor th
at affects the individual's propensity to both consume tobacco as well as
to develop the disease). Given P\, the observed probability distribution
of the pair (X\, Y)\, we ask: what is the causal effect of X on Y? In par
ticular\, can it be determined solely from the knowledge of the "correlati
on" between X and Y? In the early 1990s\, Judea Pearl proposed a formaliz
ation of the above question in the language of directed graphical models.
Put differently\, Pearl's framework formalized what it might mean to com
pute the strength of a causal relationship (such as may be measured in a c
ontrolled experiment) given only data about correlations among different c
omponents on the system.\n\nIt is intuitively clear that this problem is n
ot always solvable: the example considered above is a case in point. Howe
ver\, a long line of work by several researchers culminated in 2006 in a c
omplete algorithmic characterization of graphical models in which the prob
lem is solvable. Further\, this characterization also included an algorith
mic procedure which takes the observed distribution and outputs the requis
ite "causal" distribution in those cases where the problem is solvable [Hu
ang and Valtorta\, 2006 and Shpitser and Pearl\, 2006].\n\nThis talk will
introduce directed graphical models and causal inference problem\, and the
n give an overview of the solution of Huang-Valtorta and Shpitser-Pearl.
We will then look at some recent progress on analyzing the 'robustness' (o
r 'condition number') of these solutions with respect to "noise" or "imper
fections" in the description of the model or in the measurement of the obs
erved distribution. Surprisingly\, even though causal inference is a stat
istical problem\, such robustness questions were only asked (and partly an
swered) recently\, and several exciting future directions remain open. Ti
me permitting\, we will then look at other related notions of causal infer
ence\, such as the notion of directed mutual information in information th
eory\, and the theory of linear structural equations.\n\nDisclaimer: Part
of this is joint work with Leonard J. Schulman (available from http://www.
tifr.res.in/~piyush.srivastava/research.html#causal-inference). Some other
parts are served with gentle dollops of caveats emptor.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=5452
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5452
END:VEVENT
BEGIN:VEVENT
SUMMARY:New Approximation Algorithms for Dynamic Matching and Vertex Cover
DTSTART;VALUE=DATE-TIME:20170131T103000Z
DTEND;VALUE=DATE-TIME:20170131T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5446@cern.ch
DESCRIPTION:We consider the problem of maintaining a matching and a vertex
cover of approximately maximum (resp. minimum) size in a dynamic graph un
der a sequence of edge insertions and deletions. In recent years\, a signi
ficant body of work has been devoted to this problem. Using a primal-dual
framework\, we first show how to maintain a (2+\\epsilon) approximate maxi
mum fractional matching and minimum vertex cover in O(log n/\\epsilon^2) a
mortized update time\, where n is the number of nodes in the input graph.
Next\, we show how to perform ``deterministic rounding'' efficiently in a
dynamic setting\, thereby converting the fractional matching maintained by
the previous algorithm into an integral matching. Finally\, we show how t
o make the update time of our fractional matching algorithm worst case\, t
hereby getting the first non-trivial dynamic algorithm that is determinist
ic and has O(poly log n) worst case update time. We conclude by pointing o
ut some interesting open problems in this area (joint work with M. Henzing
er and G. Italiano (SODA 2015) and M. Henzinger and D. Nanongkai (STOC 201
6\, SODA 2017)).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=5446
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5446
END:VEVENT
BEGIN:VEVENT
SUMMARY:'Capacity' of Cellular Wireless Networks
DTSTART;VALUE=DATE-TIME:20170207T103000Z
DTEND;VALUE=DATE-TIME:20170207T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5458@cern.ch
DESCRIPTION:In a breakthrough\, we are able to give a simple description o
f 'capacity' of cellular wireless networks. Shannon capacity is intractabl
e for most wireless networks\, and a surrogate definition is used. The cap
acity is defined to be the largest ratio of the basestation density and th
e expected delay (number of retransmissions needed till success) seen at a
ny mobile. This capacity definition captures the inherent practical aspect
s of wireless networks while still being analytically non-trivial (this is
joint work with Srikanth Iyer at IISc).\n\nhttps://indico.tifr.res.in/ind
ico/conferenceDisplay.py?confId=5458
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5458
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fast Subset Convolution for Exact Exponential Algorithms
DTSTART;VALUE=DATE-TIME:20170210T103000Z
DTEND;VALUE=DATE-TIME:20170210T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5489@cern.ch
DESCRIPTION:I will present a technique know as fast subset convolution for
Exact Exponential Algorithms. We will see how this technique immediately
gives algorithms for a bunch of coloring and covering problems on graphs\,
and in particular solves the problem of finding the chromatic number of a
graph in O*(2^n) time\, which was open till 2006. (Björklund and Husfeld
t\, FOCS 2006)\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?c
onfId=5489
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5489
END:VEVENT
BEGIN:VEVENT
SUMMARY:How Limited Interaction Hinders Real Communication (and What it Me
ans for Proof and Circuit Complexity)
DTSTART;VALUE=DATE-TIME:20170214T103000Z
DTEND;VALUE=DATE-TIME:20170214T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5479@cern.ch
DESCRIPTION:Nowadays SAT solvers are able to solve problems with millions
of variables\, but some instances are still hard. While some formulas just
require large time or memory to solve\, other formulas allow to trade the
se resources. In the language of proof complexity\, they have both short p
roofs and proofs in small space\, but optimizing either measure blows up t
he other.\n\nIn this talk we will look at size-space trade-offs that hold
not only for the resolution proof system\, which captures most current sol
vers\, but for polynomial calculus and cutting planes\, which capture alge
braic and pseudo-boolean reasoning respectively. The proof goes through co
mmunication complexity\, and a key insight is to use a model of communicat
ion that captures short cutting planes proofs: communication with limited
rounds and with real numbers (joint work with Susanna F. de Rezende and Ja
kob Nordström).\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=5479
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5479
END:VEVENT
BEGIN:VEVENT
SUMMARY:Alternate Characterisations of Compact Spaces
DTSTART;VALUE=DATE-TIME:20170217T093000Z
DTEND;VALUE=DATE-TIME:20170217T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5511@cern.ch
DESCRIPTION:The usual idea of compactness of a space is that if the space
has an open cover then it has a finite subcover which is not very intuitiv
e. In this talk we attempt to look at some equivalent characterisations of
compact spaces. At first we prove some important connections between comp
actness \, totally boundedness and sequential compactness using a series o
f theorems that begin by proving that every compact metric space has the
Bolzano Weierstrass property\, move on to the Lebesgue covering lemma and
then end by proving that every sequentially compact metric space is totall
y bounded and is compact.\n\nThe second part of the talk attempts to apply
these ideas to a proof of Ascolis theorem the statement of which is given
: If $X$ is a a compact metric space then a closed subspace of $C(X\,R) o
r C(X\,C)$ is compact iff it is bounded and equicontinuous. Along the way
. We develop certain important ideas such as a metric space is compact iff
it is complete and totally bounded. We conclude by mentioning some import
ant applications of Ascolis theorem\, for example Prokhorov's theorem in p
robability.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=5511
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5511
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Generalized Method for Proving Polynomial Calculus Degree Lower
Bounds
DTSTART;VALUE=DATE-TIME:20170221T103000Z
DTEND;VALUE=DATE-TIME:20170221T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5471@cern.ch
DESCRIPTION:We study the problem of certifying unsatisfiability of formula
s in propositional logic. For proof systems such as resolution and polynom
ial calculus it is known that if the clause-variable incidence graph of a
CNF formula is an expander (i.e.\, is very well-connected)\, then proving
that this formula is unsatisfiable is hard. We further develop techniques
in [Alekhnovich and Razborov '03] to show that if one can "cluster" clause
s and variables in a way that "respects the structure" of the formula in a
certain sense\, then it is sufficient that the incidence graph of this cl
ustered version is an expander. We also give a unified view of resolution
and polynomial calculus lower bounds in terms of a 2-player game played on
this graph\, where the difference between resolution and polynomial calcu
lus is just in which player has to go first. As a corollary\, we prove tha
t the functional pigeonhole principle (FPHP) formulas are hard for polynom
ial calculus\, answering an open question in [Razborov '02].\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=5471
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5471
END:VEVENT
BEGIN:VEVENT
SUMMARY:Multiplayer Parallel Repetition for Expanding Games
DTSTART;VALUE=DATE-TIME:20170224T103000Z
DTEND;VALUE=DATE-TIME:20170224T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5535@cern.ch
DESCRIPTION:A two-player game is an important construct used in proving ma
ny hardness of approximation results. It consists of two non-communicating
players\, Alice and Bob\, trying to win against a verifier $V$\, who draw
s a question pair $(x\,y) \\in X \\times Y$ from a known distribution $D$
and sends $x$ to Alice and $y$ to Bob. The goal of Alice and Bob is to co
me up with strategies to provide answers ($a(x)\, b(y)$ resp.) to these qu
estions\, in order to win (decided by a predicate V(x\,y\,a\,b)) with maxi
mum probability over $D$. This maximum probability is called the value o
f the game.\n\nRaz first showed that repeating a two-player game in parall
el $n$-times (where $n$ question pairs are drawn independently and given t
o the players simultaneously) drives down the probability of the players w
inning all the rounds exponentially with $n$. A series of subsequent works
improved the parameters involved\, and current known results are near-opt
imal.\n\nIn contrast to two-player games\, very little is known about the
parallel-repetition of games with $k$ players for $k\\geq 3$. The only kn
own universal upper bound on the value of a $n$-repeated\, $k$-player gam
e is due to Verbitsky\; it shows a weak inverse-Ackermann decay with rega
rds to $n$. Some special classes of multi-player games (free games and an
chored games) have been shown to exhibit exponential decay in value. The
technical roadblock in extending known proofs for $k=2$ to $k \\geq 3$ i
s similar to one encountered in proving direct product results in communic
ation complexity with 3 or more players.\n\nIn this work\, we show that un
der $n$-fold repetition\, a large class of $k$-player games do\, in fact\,
exhibit an exponential decay in value. These games are expanding in a spe
cific sense. Our result recovers exponential decay theorems for free and a
nchored games as a corollary. We also point out a simple game not handled
by the above class\, and conjecture that it is in fact\, the hardest case
(oint work with Irit Dinur\, Prahladh Harsha and Henry Yuen).\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=5535
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5535
END:VEVENT
BEGIN:VEVENT
SUMMARY:Learning Inhibition Graphs Through Boolean Compressive Sensing
DTSTART;VALUE=DATE-TIME:20170303T100000Z
DTEND;VALUE=DATE-TIME:20170303T110000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5541@cern.ch
DESCRIPTION:Design of efficient sampling methodologies to capture the info
rmation content in sparse\, and thus\, "compressible" signals is well know
n as compressive sensing (CS) in the signal processing community. These ef
ficient sampling methods that use linear measurements (tests) of signals a
re known to have significantly low sampling cost compared to naive samplin
g. An analogous\, albeit a non-linear paradigm\, called group testing emer
ged much before the compressive sensing era. Boolean Compressive Sensing (
BCS) a.k.a Group Testing was introduced by Dorfman in 1943 to identify a s
mall number of unhealthy individuals (i.e.\, defective "items") amidst a l
arge population. In the BCS model\, the test outcomes are a non-linear fun
ction of the input\, and the input and the output vectors are binary. The
addition and multiplication operations in CS are replaced by Boolean-OR an
d Boolean-AND operations. The BCS framework has since been used in diverse
fields like Bioinformatics\, Theoretical Computer Science\, Computer Netw
orks\, Sports Analytics\, and Recreational Mathematics.\n\nThis talk will
focus on a newly introduced variation of the classical BCS framework that
will model a unified problem of pathogen identification and drug discovery
. The expression of a pathogen carrying item is inhibited by blocker items
in a "pooled" test. Such an activity relationship between the items is mo
delled as a bipartite graph\, called an inhibition graph. The goal is to l
earn this unknown inhibition graph with least sampling complexity. Upper b
ounds on the sampling complexity and tight non-trivial lower bounds will b
e discussed for learning the inhibition graph through "non-adaptive" BCS.\
n\nThe talk will also highlight a few other topics that I have worked on a
nd my future research plans.\n\nBiography: Abhinav Ganesan is a Senior Sys
tems Engineer at Qualcomm R&D\, Bangalore\, working on LTE system design f
or IoT applications. He was a Post Doctoral Fellow at the Institute of Net
work Coding\, The Chinese University of Hong Kong\, during 2014-2015. He o
btained his PhD from the ECE Department at the Indian Institute of Science
in 2014 and his B.E. degree from the College of Engineering\, Guindy\, An
na University in 2009. He was a recipient of the Best Academic Paper Award
at the IEEE Wireless Communications and Networking Conference (WCNC) in 2
011.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5541
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5541
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fault Tolerant Reachability\, DFS Tree\, and Min-cut
DTSTART;VALUE=DATE-TIME:20170316T083000Z
DTEND;VALUE=DATE-TIME:20170316T093000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5536@cern.ch
DESCRIPTION:Consider the following problem of single source reachability u
nder failures of vertices or edges. Let $G$ be a given directed graph on n
vertices with a designated source vertex $s$ \, and $k$ be any positive i
nteger. Compute the sparsest subgraph of $G$ that preserves reachability
from s upon failure of any k vertices/edges.\n\nIn this talk\, we shall di
scuss two algorithms that construct such a subgraph based on respectively
DFS tree and min-cut: two elementary and distinct concepts seemingly unrel
ated to the problem. The first algorithm\, that works only for the special
case of $k = 1$\, is based on a DFS tree rooted at $s$ . This solution ca
n be seen as a byproduct of the seminal work on dominators by Lengeuer and
Tarjan in 1979. The corresponding subgraph will always have less than 2$n
$ edges. The algorithm for any general value of $k$ was derived after 35 y
ears. This algorithm turns out to be based on farthest min-cut : a term co
ined by Ford and Fulkerson in their 1962 research paper on the first algor
ithm for max-flow. The corresponding subgraph will have $O(n2k)$ edges. We
also prove a matching lower bound (this is a joint work with Keerti Choud
hary (a PhD student at IITK) and Liam Roditty (Bar-Ilan University)).\n\nh
ttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5536
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5536
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum De-Finetti Representation Theorem for Infinite Exchangeabl
e Probability Assignments
DTSTART;VALUE=DATE-TIME:20170317T114500Z
DTEND;VALUE=DATE-TIME:20170317T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5569@cern.ch
DESCRIPTION:In this talk\, we will discuss De Finetti representation theor
em on exchangeable probability assignment\, that provides an operational d
efinition of the concept of an unknown probability in Bayesian probability
theory\, where probabilities are taken to be degrees of belief instead of
objective states of nature.\n\nl give a brief description of classical ve
rsion and then move onto the quantum analogue.\n\nl discuss the proof of t
he theorem in quantum setting\, given by Fuchs et. al.\n\nThere are wide a
pplications of this theorem\, mainly in tomography\, where a probability d
istribution (a prior) is to be estimated by repeated trials on identically
prepared systems.\n\nl cover the background required in quantum setting a
s we progress in the talk.\n\nhttps://indico.tifr.res.in/indico/conference
Display.py?confId=5569
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5569
END:VEVENT
BEGIN:VEVENT
SUMMARY:Adaptive Matching for Expert Systems With Uncertain Task Types
DTSTART;VALUE=DATE-TIME:20170320T103000Z
DTEND;VALUE=DATE-TIME:20170320T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5568@cern.ch
DESCRIPTION:Online two-sided matching markets such as Q&A forums (e.g. Sta
ckOverflow\, Quora) and online labour platforms (e.g. Upwork) critically r
ely on the ability to propose adequate matches based on imperfect knowledg
e of the two parties to be matched. This prompts the following question: W
hich matching recommendation algorithms can\, in the presence of such unce
rtainty\, lead to efficient platform operation?\n\nTo answer this question
\, we develop a model of a task / server matching system. For this model\,
we give a necessary and sufficient condition for an incoming stream of ta
sks to be manageable by the system. We further identify a so-called back-p
ressure policy under which the throughput that the system can handle is op
timized. We show that this policy achieves strictly larger throughput than
a natural greedy policy. Finally\, we validate our model and confirm our
theoretical findings with experiments based on logs of Math.StackExchange\
, a StackOverflow forum dedicated to mathematics (joint work with Lennart
Gulikers\, Laurent Massoulie\, and Milan Vojnovic).\n\nBio:Virag Shah is c
urrently a Postdoctoral Researcher at Microsoft Research - Inria Joint cen
ter\, Palaiseau\, France. His research interests include design of online
two-sided matching markets and of networked systems for performance. He ob
tained his PhD from ECE department at The University of Texas at Austin\,
USA\, and M.E. from ECE department at IISc\, Bangalore. He has received Be
st Paper awards at IEEE INFOCOM 2014 and NCC 2010.\n\nhttps://indico.tifr.
res.in/indico/conferenceDisplay.py?confId=5568
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5568
END:VEVENT
BEGIN:VEVENT
SUMMARY:Universal Lossless Compression of Graph-indexed Data
DTSTART;VALUE=DATE-TIME:20170330T103000Z
DTEND;VALUE=DATE-TIME:20170330T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5584@cern.ch
DESCRIPTION:Many modern data sources arising from social networks\, biolog
ical data\, etc. are best viewed as indexed by combinatorial structures su
ch as graphs\, rather than as time series. The local weak limit theory for
sparse graphs\, also known as the objective method\, provides a framework
in which to make sense of what one might mean by a stationary stochastic
process indexed by graphs. The theory of time series is the engine driving
an enormous range of applications in areas such as control theory\, commu
nications\, information theory and signal processing\, to name just a few
elds. It is to be expected that a theory of stationary stochastic processe
s indexed by combinatorial structures\, in particular graphs\, would event
ually have a similarly wide ranging impact.\n\nConsider a graph whose vert
ices and edges carry marks\, viewed as data samples associated to the vert
ex or the edge respectively. For example the vertices might be individuals
\, the vertex mark might describe the musical taste of the individual and
the edge mark might describe the degree of anity between the corresponding
pair of individuals. We pose the problem of representing the information
content of such a marked graph in the most ecient way possible. Further\,
we would like to do this in a universal sense\, i.e.\, in eect\, without m
aking any statistical assumptions about the data sample being compressed.
It turns out that one can make precise sense of this question in the langu
age of the local weak limit theory\, the only assumption being that the gr
aph is sparse\, i.e. that the number of edges is on the same scale as the
number of vertices. What is more\, the compression can be done in the most
ecient way possible. Namely one can make sense of a notion of entropy ass
ociated to the local weak limit\, which is on a linear scale in the number
of nodes\, and one can compress down to this entropy in a universal sense
. The entropy notion we work with is a generalized version of the one intr
oduced by Bordenave and Caputo (2014).\n\nAll the necessary background fro
m the theory of local weak limits that is needed will be developed during
the talk (joint work with Payam Delgosha).\n\nhttps://indico.tifr.res.in/i
ndico/conferenceDisplay.py?confId=5584
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5584
END:VEVENT
BEGIN:VEVENT
SUMMARY:Large Sample Behaviour of High Dimensional Autocovariance Matrices
with Application
DTSTART;VALUE=DATE-TIME:20170331T093000Z
DTEND;VALUE=DATE-TIME:20170331T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5586@cern.ch
DESCRIPTION:Consider a sample of size $n$ from a linear process of dimensi
on $p$ where $n\, p \\to \\infty$\, $p/n \\to y \\in [0\, \\ \\infty).$ Le
t $\\hat{\\Gamma}_{u}$ be the sample autocovariance of order $u$.\n\nThe e
xistence of limiting spectral distribution (LSD) of $\\hat{\\Gamma}_{u} +
\\hat{\\Gamma}_{u}^{*}$\, the symmetric sum of the sample autocovariance m
atrix $\\hat{\\Gamma}_{u}$ of order $u$\, is known in the literature under
appropriate (strong) assumptions on the coefficient matrices. Under signi
ficantly weaker conditions\, we prove\, in a unified way\, that the LSD of
any symmetric polynomial in these matrices such as $\\hat{\\Gamma}_{u} +
\\hat{\\Gamma}_{u}^{*}$\, $\\hat{\\Gamma}_{u}\\hat{\\Gamma}_{u}^{*}$\, $\\
hat{\\Gamma}_{u \\hat{\\Gamma}_{u}^{*}+\\hat{\\Gamma}_{k}\\hat{\\Gamma}_{k
}^{*}$ exist.\n\nOur approach is through the more intuitive algebraic meth
od of free probability that is applicable after an appropriate embedding\,
in conjunction with the method of moments. Thus\, we are able to provide
a general description for the limits in terms of some freely independent v
ariables. All the previous results followas special cases.\n\nWe suggest s
tatistical uses of these LSD and related results in problems such as order
determination and white noise testing.\n\nhttps://indico.tifr.res.in/indi
co/conferenceDisplay.py?confId=5586
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5586
END:VEVENT
BEGIN:VEVENT
SUMMARY:Continuous-time Principal-Agent Problem: A Stackelberg Stochastic
Differential Game
DTSTART;VALUE=DATE-TIME:20170404T103000Z
DTEND;VALUE=DATE-TIME:20170404T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5488@cern.ch
DESCRIPTION:We provide a systematic method for solving general Principal-A
gent problems. Our main result reduces such Stackelberg stochastic differe
ntial games to a standard stochastic control problem\, which may be addres
sed by the standard tools of control theory. Our proofs rely on the backwa
rd stochastic differential equations approach to non-Markovian stochastic
control\, and more specifically\, on the recent extensions to the second o
rder case.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=5488
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5488
END:VEVENT
BEGIN:VEVENT
SUMMARY:Secure Message Transmission: Tips and Tricks
DTSTART;VALUE=DATE-TIME:20170407T114500Z
DTEND;VALUE=DATE-TIME:20170407T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5611@cern.ch
DESCRIPTION:Suppose A and B are parties in a network and want to communica
te with each other privately. This problem is trivial if A and B have a pr
ivate communication link between them. What if there is no such link? This
is the problem of secure message transmission (SMT). We consider the char
acterization of networks in which SMT is possible when the adversary is se
mi-honest or malicious and when security required is perfect or statistica
l etc. Some of the protocols are interesting in their own right and some o
f the techniques used for proving the impossibilities are useful for provi
ng impossibility results generally in secure multi party computing.\n\nhtt
ps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5611
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5611
END:VEVENT
BEGIN:VEVENT
SUMMARY:Generating Correlated Random Variables via Shared Randomness
DTSTART;VALUE=DATE-TIME:20170411T103000Z
DTEND;VALUE=DATE-TIME:20170411T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5613@cern.ch
DESCRIPTION:In coordination problems\, agents have to perform actions so t
hat their joint actions produce outputs according to some prescribed joint
distribution. Such problems arise naturally in several diverse areas such
as control\, game theory and task assignment. An early work of this kind
in information theory literature is due to Wyner(1975) who characterized t
he minimum rate of common randomness required by two agents to sample appr
oximately from the joint distribution $p(x\,y)$ of correlated random varia
bles $X\,Y$. In practice\, shared randomness is also an important aspect o
f interest. We consider a problem in which in addition to Wyner's setting\
, the server has access to two independent sources of randomness each of w
hich is shared with a different agent. The goal is to minimize the rate of
common message. We present some results in relation to this.\n\nhttps://i
ndico.tifr.res.in/indico/conferenceDisplay.py?confId=5613
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5613
END:VEVENT
BEGIN:VEVENT
SUMMARY:Dual Polynomials and Communication Complexity of XOR Functions
DTSTART;VALUE=DATE-TIME:20170414T103000Z
DTEND;VALUE=DATE-TIME:20170414T120000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5626@cern.ch
DESCRIPTION:In this talk\, we will see how certain "degree hardness" prope
rties of a function (e.g approximate degree\, sign-degree) amplify to give
us "monomial based" hardness properties of a certain "lift" (as defined b
y Krause and Pudlak (STOC '94)) of the function. We list applications to s
ymmetric functions\, resolving conjectures of Ada et al. (APPROX-RANDOM '1
2) and Zhang ('91).\n\nWe will also see a connection between the polynomia
l margin of a function f\, and the discrepancy of f of XOR via linear prog
ramming duality. Using this duality\, we demonstrate polynomial based tech
niques for understanding the bounded error communication complexity and we
akly-unbounded error communication complexity of XOR functions. This allow
s us to prove a weak form of a conjecture by Zhang and Shi (Quantum Inform
ation and Computation '09) and reproves a result of Goldmann et al. (Compu
tational Complexity '92) (based on joint work with Arkadev Chattopadhyay (
https://arxiv.org/abs/1704.02537)).\n\nhttps://indico.tifr.res.in/indico/c
onferenceDisplay.py?confId=5626
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5626
END:VEVENT
BEGIN:VEVENT
SUMMARY:Computational Studies of the Indus Script
DTSTART;VALUE=DATE-TIME:20170419T103000Z
DTEND;VALUE=DATE-TIME:20170419T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5620@cern.ch
DESCRIPTION:Indus script is a major undeciphered script. Several attempts
have been made to decipher the script but there is no consensus about its
content. The lack of definite knowledge about its structure makes it diffi
cult to objectively evaluate any claim of decipherment. We have tried to f
ill this lacuna by analyzing the syntax of the script in detail. The focus
of our study is to identify patterns in the Indus writing and explore its
underlying logic without making any assumptions about its content. In the
present talk\, I will give an overview of our work on the structure of th
e Indus script using various techniques related to computational linguisti
cs.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5620
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5620
END:VEVENT
BEGIN:VEVENT
SUMMARY:On the Communication and Streaming Complexity of Maximum Bipartite
Matching
DTSTART;VALUE=DATE-TIME:20170421T114500Z
DTEND;VALUE=DATE-TIME:20170421T131500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5639@cern.ch
DESCRIPTION:Consider the following communication problem. Alice holds a gr
aph $G_A = (P\, Q\, E_A)$ and Bob holds a graph $G_B = (P\, Q\, E_B)$\, wh
ere $|P| = |Q| = n$. Alice is allowed to send Bob a message $m$ that depen
ds only on the graph $G_A$. Bob must then output a matching $M\\subseteq E
_A \\cup E_B$. What is the minimum message size of the message $m$ that Al
ice sends to Bob that allows Bob to recover a matching of size at least $(
1 - \\epsilon)$ times the maximum matching in $G_A \\cup G_B$? The minimum
message length is the one-round communication complexity of approximating
bipartite matching. It is easy to see that the one-round communication co
mplexity also gives a lower bound on the space needed by a one-pass stream
ing algorithm to compute a $(1 - \\epsilon)$-approximate bipartite matchin
g. Using connection to $\\epsilon$-RS graphs\, the authors show that for a
ny $\\delta>0$\, a $(2/3+\\delta)$-approximation requires a communication
complexity of $n^{1+\\Omega(1/\\log\\log n)}$. And thus\, no one-pass semi
-streaming algorithm can have a $(2/3+\\delta)$-approximation ratio (autho
rs: Ashish Goel\, Michael Kapralov\, Sanjeev Khanna).\n\nhttps://indico.ti
fr.res.in/indico/conferenceDisplay.py?confId=5639
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5639
END:VEVENT
BEGIN:VEVENT
SUMMARY:Content Caching and Delivery with Partial Adaptive Matching
DTSTART;VALUE=DATE-TIME:20170425T103000Z
DTEND;VALUE=DATE-TIME:20170425T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5467@cern.ch
DESCRIPTION:Caching of popular content during off-peak hours is a strategy
to reduce the network load during peak hours. We consider a model where m
ultiple caches store pre-fetched content and when users request files\, th
ey are matched to caches based on the request pattern. In particular\, we
focus on the case where caches are divided into clusters and each user can
only be assigned to a unique cache from a specific cluster. This is a gen
eralization of two popular models which are the extremes of the proposed m
odel: one where each user is pre-attached to a cache irrespective of what
it demands (static matching) and the other where each user can be assigned
to any cache in the entire network (fully adaptive matching). We show th
at neither the coded delivery strategy (approximately optimal when the use
r-cache assignment is pre-fixed) nor the uncoded replication strategy (app
roximately optimal when all caches belong to a single cluster) is sufficie
nt for all memory regimes. We propose a hybrid solution that combines idea
s from both schemes and that performs strictly better than both. Finally\,
we show that this hybrid strategy is approximately optimal in most memory
regimes.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId
=5467
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5467
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Mathematical Journey Into Human-centric Networks and Their Inter
disciplinary Connections
DTSTART;VALUE=DATE-TIME:20170428T063000Z
DTEND;VALUE=DATE-TIME:20170428T073000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5588@cern.ch
DESCRIPTION:Many networks and platforms centered around online populations
have flourished in the last few decades\, riding on the rapid proliferati
on of web technologies. These\, collectively known as human-centric networ
ks\, have various objectives like social interaction\, commerce\, content
sharing\, and advertising. Among them\, online social networks have unleas
hed the potential of the web as a medium for information sharing and have
had significant social and economic impact. Crowdsourcing systems\, which
are another emerging and popular kind of online platform\, are giving a n
ew perspective on the web as a huge work force of diversely-skilled people
. Particularly\, non-profit impact sourcing platforms which train and empl
oy underprivileged people to do computationally intractable tasks like spe
ech transcription and handwriting recognition are of high relevance to cou
ntries like India that are rich in human resources. I shall discuss mathe
matical problems motivated by practice\, including on opinion dynamics in
social networks\, task allocation in crowdsourcing\, and overloading of hu
man workers. Reliable computation using unreliable units is a big challen
ge in crowdsourcing because people make errors\; this is also a major chal
lenge in emerging nanoscale circuits made of spin devices and carbon nanot
ubes\, and also in cyber-physical systems\, internet of things\, and infra
structure networks. I shall discuss applications of information network-ba
sed theories in nanoscale circuits. I shall conclude my talk with some di
scussion on future research directions in social networks\, crowdsourcing\
, and nanoscale circuits\, as well as their connections.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=5588
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5588
END:VEVENT
BEGIN:VEVENT
SUMMARY:Insensitivity of the Mean Field of Loss Systems Under Randomized S
Q(d) Algorithms
DTSTART;VALUE=DATE-TIME:20170502T103000Z
DTEND;VALUE=DATE-TIME:20170502T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5554@cern.ch
DESCRIPTION:In many applications such as cloud computing\, managing server
farm resources etc. an incoming task or job has to be matched with an app
ropriate server in order to minimise the latency or blocking associated wi
th the processing. Ideally the best choice would be to match a job to the
fastest available server. However when there are thousands of servers requ
iring the information on all server tasks is an overkill.\n\nPioneered in
the 1990's the idea of randomised sampling of a few servers was proposed b
y Vvedenskaya and Dobrushin in Russia and Mitzmenmacher in the US and popu
larised as the "power of two" schemes which basically means that sampling
two servers randomly and sending the job to the "better" server (i.e. with
the shortest queue\, or most resources) provides most of the benets of sa
mpling all the servers.\n\nIn the talk I will discuss multi-server loss mo
dels under power-of-d routing scheme when service time distributions are g
eneral with nite mean. Previous works on these models assume that the serv
ice times are exponentially distributed and insensitivity was suggested th
rough simulations. Showing insensitivity to service time distributions has
remained an open problem. The diculty is that for general service times t
he underlying Markovian model is more complex. Using a measure valued proc
ess approach we rst derive the mean eld equation (MFE) for the empirical m
easure. The MFE is now characterized by a pde whose stationary point coinc
ides with the xed point in the case with exponential service times. This e
stablishes the insensitivity of the xed point. The techniques developed in
this paper are applicable to study mean eld limits for Markov processes o
n general state spaces and insensitivity properties of other queueing mode
ls.\n\nBio: The speaker was educated at the Indian Institute of Technology
\, Bombay (B.Tech\, 1977)\, Imperial College\, London (MSc\, DIC\, 1978) a
nd obtained his PhD under A. V. Balakrishnan at UCLA in 1983. He is curren
tly a University Research Chair Professor in the Dept. of ECE at the Unive
rsity of Waterloo\, Ont.\, Canada where he has been since September 2004.
Prior to this he was Professor of ECE at Purdue University\, West Lafayett
e\, USA. He is a D.J. Gandhi Distinguished Visiting Professor at the India
n Institute of Technology\, Bombay. He is a Fellow of the IEEE and the Roy
al Statistical Society. He is a recipient of the Best Paper Awards at INFO
COM 2006\, the International Teletrac Congress 2015\, Performance 2015\, a
nd was runner-up for the Best Paper Award at INFOCOM 1998. His research in
terests are in modeling\, control\, and performance analysis of both wirel
ine and wireless networks\, and in applied probability and stochastic anal
ysis with applications to queueing\, ltering\, and optimization.\n\nhttps:
//indico.tifr.res.in/indico/conferenceDisplay.py?confId=5554
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5554
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Leonid Gurvits’s Proof for Van der Waerden Conjecture
DTSTART;VALUE=DATE-TIME:20170505T114500Z
DTEND;VALUE=DATE-TIME:20170505T131500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5667@cern.ch
DESCRIPTION:The Van der Waerden Conjecture states that the permanent of a
doubly stochastic matrix n x n matrix is at least n!/n^n\, which is the c
ase when each entry of the matrix is $1/n$. Though this conjecture is simp
le to state\, it was unsolved for over fifty years until it was proved by
Falikman (1979) and Egorychev (1980). In 2008\, Leonid Gurvits came up wit
h an amazingly short proof of the Van der Waerden Conjecture using H-stabl
e polynomials. This proof will be the primary focus of this talk.\n\nThe p
roof is self-contained\, and we will follow the exposition by Laurent and
Schrijver.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=5667
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5667
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stochastic Dominance (SD) and its Application to Enhanced Indexati
on
DTSTART;VALUE=DATE-TIME:20170512T114500Z
DTEND;VALUE=DATE-TIME:20170512T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5673@cern.ch
DESCRIPTION:An individual is faced with several options while deciding to
invest in a financial market. Some opportunities giving higher returns but
carrying more risk while some being less risky and give lower returns. Ea
ch individual needs to rank the investment opportunities according to his
preference. Stochastic Dominance (SD) is a method for ranking the random v
ariables. In particular\, second-order stochastic dominance (SSD) is relat
ed to the risk-averse investor preferences. In this talk\, I will introduc
e the concept of SD. We will also look into an application of SSD (for a r
isk-averse investor) to construct a portfolio that generates higher return
s as compared to the reference index.\n\nhttps://indico.tifr.res.in/indico
/conferenceDisplay.py?confId=5673
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5673
END:VEVENT
BEGIN:VEVENT
SUMMARY:Time-Space Lowerbound for SAT
DTSTART;VALUE=DATE-TIME:20170519T114500Z
DTEND;VALUE=DATE-TIME:20170519T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5681@cern.ch
DESCRIPTION:In this talk\, we look at a $n^1.6616$ lower bound for SAT on
$n^{o(n)}$ space machines. The result is taken from the 2006 paper by Ryan
Williams.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confI
d=5681
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5681
END:VEVENT
BEGIN:VEVENT
SUMMARY:Using Git for Writing Papers
DTSTART;VALUE=DATE-TIME:20170526T114500Z
DTEND;VALUE=DATE-TIME:20170526T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5690@cern.ch
DESCRIPTION:uring academic collaborations\, we have different people worki
ng on a paper and the draft has a natural evolution. Most people are ok us
ing a sync service like dropbox\, or even just emailing tex sources to eac
h other (gasp). Similar evolutions of projects also happen in the software
industry\, where the are dealing with millions of lines of code and thous
ands of developers contributing stuff. And they manage their code via some
version control system. This talk is to introduce a popular version contr
ol system called ‘git’\, and the talk is going to be specifically tail
ored towards using git for writing papers etc. Mainly\, it is to convince
you guys that git is awesome\, and easy\, and really makes sense for acade
mic papers.\n\nAs a disclaimer\, the talk would *not* be about the *right*
way to use ‘git’ but rather what in my opinion is the ‘easiest’ w
ay to use git. I will introduce the basic ideas\, and much of the talk wou
ld just be a demo of the various things you can do with git.\n\nRequest to
the audience: If you have your laptop\, please bring it during the talk s
o that you can try out the examples by yourself\; it would really accelera
te the learning. Preferably\, you can install git before the talk (For ubu
ntu/debian\, you probably have to run ‘sudo apt-get install git’ on a
terminal. For Windows\, you can install from via www.git-scm.com etc.) Thi
s would then let me focus the talk on just ‘git’ rather than OS-specif
ic installation processes etc.\n\nhttps://indico.tifr.res.in/indico/confer
enceDisplay.py?confId=5690
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5690
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Query and Communication Complexity
DTSTART;VALUE=DATE-TIME:20170601T050000Z
DTEND;VALUE=DATE-TIME:20170601T060000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5693@cern.ch
DESCRIPTION:In the first part\, we will consider the connection between tw
o well-known complexity measures --- communication complexity and query co
mplexity. For a composed function $f \\circ g$\, where $f: \\{0\,1\\}^p \\
rightarrow \\mathcal{Z}$ and $g : \\{0\,1\\}^n \\times \\{0\,1\\}^n \\righ
tarrow \\{0\,1\\}$\, we will show a sufficient condition for $g$ such that
the communication complexity of $f \\circ g$ becomes tightly bounded by t
he query complexity of $f$ times the communication complexity of $g$. Such
a theorem is known as simulation theorem or lifting theorem in literature
. We will also show two well-known functions for which the said condition
holds. This part deals with deterministic complexity measures.\n\nIn the s
econd part of the talk\, we will deal with randomized complexity measure.
We consider a specific composed function $\\text{elim} \\circ g$\, namely
the elimination function\, and we will demonstrate a key property of $g$\,
namely the regularity property\, which is closely related to the communic
ation complexity of the elimination problem. To this end\, we will show a
tight connection between regularity and another well-studied pseudo-random
property of $g$\, namely\, discrepancy.\n\nIn the third part\, we will ge
t into the realm of asymmetric communication complexity where one of the p
layers holds an input of considerably smaller length than that of the othe
r player. We will show two results here --- (a) We will consider function
like $g^p$ and show direct-sum results for such functions in asymmetric se
tting\, and (b) we will show tight randomized lower bound for a composed f
unction\, namely\, the asymmetric unordered search function. We will allud
e to the role of regularity in obtaining such a lower bound.\n\nIn the las
t part of the talk\, we will talk about communication complexity in multip
arty number-in-hand coordinator model. We will show a tight lower bound fo
r a well-studied function (in two-party model)\, namely the tribes functio
n\, in this setting.\n\nhttps://indico.tifr.res.in/indico/conferenceDispla
y.py?confId=5693
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5693
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Snake Lemma
DTSTART;VALUE=DATE-TIME:20170602T114500Z
DTEND;VALUE=DATE-TIME:20170602T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5697@cern.ch
DESCRIPTION:In this talk\, we will see a proof of the snake lemma. The tal
k will assume a basic familiarity (group homomorphisms\, kernels\, cokerne
ls\, etc.) with group theory. Time permitting\, the speaker will also expl
ain an application of the snake lemma to his own work.\n\nhttps://indico.t
ifr.res.in/indico/conferenceDisplay.py?confId=5697
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5697
END:VEVENT
BEGIN:VEVENT
SUMMARY:Modelling Uncertainty: A Tale That Tails Tell
DTSTART;VALUE=DATE-TIME:20170606T053000Z
DTEND;VALUE=DATE-TIME:20170606T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5669@cern.ch
DESCRIPTION:In this non-technical talk\, we discuss uncertainty – why it
needs to be modelled – the role of probability theory\, brief history o
f probability\, and some of its popular applications. We focus especially
on modelling tail events - events that occur with small probability but c
an lead to catastrophic consequences. For last two centuries\, many natura
l and social phenomenon were thought to follow the light-tailed Normal dis
tribution or the `Bell Curve’. Lately\, particularly in social settings\
, one sees fat-tailed distributions. In this talk\, we also shed light on
the underlying principles governing these distributions.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=5669
LOCATION: AG-66 (Lecture Theatre)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5669
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Composition Theorem for Randomized Query Complexity
DTSTART;VALUE=DATE-TIME:20170609T114500Z
DTEND;VALUE=DATE-TIME:20170609T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5704@cern.ch
DESCRIPTION:Around a week ago\, a group of researchers* proved that the ra
ndomized query complexity of f composed with g is lower bounded by the pro
duct of the randomized query complexities of f and g\, albeit with subopti
mal parameters. This proof of theirs is quite simple and so I wish to give
a decently good exposition of it.\n\nIn this talk\, I plan to:\n\n- intro
duce query complexity\n- show Yao's lemma and how it simplifies dealing wi
th randomized algorithms\n- present the proof from the paper\n\n* Anurag A
nshu\, Dmitry Gavinsky\, Rahul Jain\, Srijita Kundu\, Troy Lee\, Priyanka
Mukhopadhyay\, Miklos Santha and Swagato Sanyal\, in their paper "A Compos
ition Theorem for Randomized Query Complexity" (arXiv:1706.00335)\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5704
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5704
END:VEVENT
BEGIN:VEVENT
SUMMARY:Exact Algorithms for Some Graph Coloring and Covering Problems
DTSTART;VALUE=DATE-TIME:20170613T053000Z
DTEND;VALUE=DATE-TIME:20170613T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5682@cern.ch
DESCRIPTION:Inclusion-Exclusion based methods will be presented for solvin
g exactly some graph coloring and covering problems. In particular this so
lves the problem of finding the Chromatic number of a graph in O*(2^n) tim
e\, which was open till 2006 (Björklund and Husfeldt\, FOCS 2006). Then\,
an alternate approach of using Fast Subset Convolution will be presented:
this generalizes the Inclusion-Exclusion based methods at a cost. Finally
\, an algorithm for computing the b-fold Fractional Chromatic number in O*
(2^(n log(2b))) time\, will be presented.\n\nhttps://indico.tifr.res.in/in
dico/conferenceDisplay.py?confId=5682
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5682
END:VEVENT
BEGIN:VEVENT
SUMMARY:How to be a Great Researcher in Computer Science?
DTSTART;VALUE=DATE-TIME:20170614T093000Z
DTEND;VALUE=DATE-TIME:20170614T103000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5668@cern.ch
DESCRIPTION:Computer science(CS) has emerged in last 40+ years to be a bra
nch of science on its own. The concerns in CS involve all the way from fun
damentals of mathematics to designing usable social networking apps. Since
CS is a relatively new science\, it has developed its own culture of doin
g research that is quite distinct than other fields of research. In this t
alk\, we will discuss key aspects of the culture and how to adapt oneself
to be a great computer scientist.\n\nDisclaimer: This talk is from the vie
w point of formal methods research. The speaker has learned the lessons pr
esented in the talk by working with other great computer scientists. He hi
mself aspiring to implement the lessons in his own work.\n\nhttps://indico
.tifr.res.in/indico/conferenceDisplay.py?confId=5668
LOCATION: AG-80
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5668
END:VEVENT
BEGIN:VEVENT
SUMMARY:Streaming Complexity of Approximating Max 2CSP and Max Acyclic Sub
graph
DTSTART;VALUE=DATE-TIME:20170616T114500Z
DTEND;VALUE=DATE-TIME:20170616T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5716@cern.ch
DESCRIPTION:This paper is joint work with Venkatesan Guruswami and Ameya V
ellingker\, to appear in APPROX 17.\n\nWe studied the complexity of estima
ting the optimum value of a Boolean 2CSP (arity two constraint satisfactio
n problem) in the single-pass streaming setting\, where the algorithm is p
resented the constraints in an arbitrary order. We give a streaming algori
thm to estimate the optimum within a factor approaching 2/5 using logarith
mic space\, with high probability. This beats the trivial factor 1/4 estim
ate obtained by simply outputting 1/4 th of the total number of constraint
s.\n\nThe inspiration for our work is a lower bound of Kapralov\, Khanna\,
and Sudan (SODA '15) who showed that a similar trivial estimate (of facto
r 1/2) is the best one can do for Max CUT. This lower bound implies that b
eating a factor 1/2 for Max 2CSP\, in particular\, to distinguish between
the case when the optimum is m/2 versus when it is at most (1/4+ \\eps) m\
, where m is the total number of constraints\, requires polynomial space.
We complement this hardness result by showing that one can distinguish bet
ween the case in which the optimum exceeds (1/2 + \\eps)m and the case in
which it is close to m/4.\n\nWe also prove that estimating the size of the
maximum acyclic subgraph of a graph\, when its edges are presented in a s
ingle-pass stream\, within a factor better than 7/8 requires polynomial sp
ace.\n\nIn this talk\, I'll present the main results of the paper. The pro
ofs are quite simple and would not require any previous knowledge\n\nhttps
://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5716
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5716
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Model of Casino Gambling Using Prospect Theory
DTSTART;VALUE=DATE-TIME:20170620T090000Z
DTEND;VALUE=DATE-TIME:20170620T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5713@cern.ch
DESCRIPTION:The first part of the talk will be an introduction to Kahneman
and Tversky's seminal Prospect Theory\, which is a descriptive theory of
how people evaluate risk. We will then see a rich model of casino gambling
\, which captures some features of actual gambling behavior (Barberis 2012
). We will analyze\, in particular\, time inconsistencies that are observ
ed in casinos : a player might enter the casino with an exit strategy in m
ind\; but once he starts playing\, he wants to switch to a different strat
egy.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5713
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5713
END:VEVENT
BEGIN:VEVENT
SUMMARY:No Title
DTSTART;VALUE=DATE-TIME:20170622T023000Z
DTEND;VALUE=DATE-TIME:20170622T033000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5723@cern.ch
DESCRIPTION:test\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py
?confId=5723
LOCATION:
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5723
END:VEVENT
BEGIN:VEVENT
SUMMARY:Algebraic Branching Programs - Nisan's Characterization and Identi
ty Testing
DTSTART;VALUE=DATE-TIME:20170623T114500Z
DTEND;VALUE=DATE-TIME:20170623T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5724@cern.ch
DESCRIPTION:Nisan gave an exact characterization of the ``non-commutative
algebraic branching program (ABP) complexity'' of a polynomial in 1991. He
showed that the size of a non-commutative ABP computing a polynomial $f$
is exactly equal to the sum of ranks of the \\emph{coefficient matrices} o
f $f$. The characterization since then has been extended to other models o
f computation\, particularly to Read once Oblivious ABPs (ROABPs)\, and ha
s motivated many interesting results.\n\nIn this talk we will first formal
ly define ROABPs and then see a proof of Nisan's characterization. As an a
pplication\, we will then proceed to see a \\emph{deterministic} identity
test for polynomials computed by ROABPs which is an extension of a work by
Raz and Shpilka from 2005. In the remaining time (if any) we will see an
overview of other models to which the characterization extends.\n\nhttps:/
/indico.tifr.res.in/indico/conferenceDisplay.py?confId=5724
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5724
END:VEVENT
BEGIN:VEVENT
SUMMARY:Universal Composability of Secure Multi-party Computation Protocol
s
DTSTART;VALUE=DATE-TIME:20170630T114500Z
DTEND;VALUE=DATE-TIME:20170630T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5729@cern.ch
DESCRIPTION:We discuss the notion of security in the multi-party computing
protocols. We start with the simpler definition of standalone security. T
he need for a stronger definition of security is motivated by a simple exa
mple where standalone security is shown to be insufficient when we compose
protocols. The rest of the discussion would be on UC security\, which is
the strongest notion of security. We discuss this concept with some exampl
es.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5729
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5729
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Sparse Recovery Framework for Fast and Efficient Network Applica
tions
DTSTART;VALUE=DATE-TIME:20170719T053000Z
DTEND;VALUE=DATE-TIME:20170719T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5756@cern.ch
DESCRIPTION:Modern communication networks present both significant challen
ges as well as opportunities that are distinct from traditional networks.
For example\, while the huge number of connected devices in applications s
uch as IoT (Internet of Things) suggests stringent bandwidth and complexit
y requirements for communication tasks\, often\, the underlying sparsity i
n the problem greatly reduces the resources needed. With the above motivat
ion in mind\, in this talk\, we present a class for sparse recovery algori
thms that are optimal or near-optimal in terms of the speed of decoding an
d the number of measurements needed.\n\nWe discuss a novel conceptual fram
ework -- "picking and peeling" -- for fast and efficient algorithms for fo
ur important sparse recovery problems -- compressive sensing\, compressive
phase retrieval\, group testing\, and network tomography. Using this prim
itive\, we begin by describing our compressive sensing algorithm SHO-FA (f
or SHOrt and FAst) which achieves a decoding complexity of O(k) while usin
g only O(k) measurements (this is the fastest possible performance in the
order sense). Next\, we will briefly describe our algorithms for the other
three problems. For each of these problems\, our algorithms are either or
der-optimal or near-optimal both in terms of two metrics - the number of m
easurements and the time complexity of decoding. Finally\, we examine anot
her metric for performance - the energy required by the VLSI circuit that
implements these algorithms. Surprisingly\, even algorithms that are effic
ient in terms of time complexity are no longer efficient in terms of decod
ing energy. We show a novel information theoretic lower bound on the decod
ing energy required in compressive sensing and briefly describe some ideas
that may allow us to approach this bound.\n\nBio: Mayank Bakshi is a Rese
arch Assistant Professor at the Institute of Network Coding\, Chinese Univ
ersity of Hong Kong. Prior to this\, he obtained his B.Tech and M.Tech fro
m IIT Kanpur in 2003 and 2005 respectively\, and PhD from Caltech in 2011\
, all in Electrical Engineering. From 2012-2014\, he was a post-doctoral f
ellow at Institute of Network Coding\, CUHK. His research interests includ
e sparse signal recovery\, information theoretic security\, and network co
ding.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=575
6
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5756
END:VEVENT
BEGIN:VEVENT
SUMMARY:Credit Risk: Simple Closed Form Approximate Maximum Likelihood Est
imator
DTSTART;VALUE=DATE-TIME:20170721T114500Z
DTEND;VALUE=DATE-TIME:20170721T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5765@cern.ch
DESCRIPTION:Discrete default intensity based or logit type models are comm
only used as reduced form models for conditional default probabilities for
corporate loans where this default probability depends upon macroeconomic
as well as firm-specific covariates. Typically\, maximum likelihood (ML)
methods are used to estimate the parameters associated with these models.
Since defaults are rare\, a large amount of data is needed for this estima
tion resulting in a computationally time consuming optimization. In this t
alk\, we observe that since the defaults are typically rare\, the first or
der equations from ML estimation suggest a simple\, accurate and intuitive
ly appealing closed form estimator of the underlying parameters. We analyz
e the properties of the proposed estimator as well as the ML estimator in
a statistical asymptotic regime where the conditional probabilities decrea
se to zero\, the number of firms as well as the data availability time per
iod increases to infinity. We characterize the dependence of the mean squa
re error of the estimator on the number of firms as well as time period of
available data. Our conclusion\, validated by numerical analysis\, is tha
t when the underlying model is correctly specified\, the proposed estimato
r is typically similar or only slightly worse than the ML estimator. Impor
tantly however\, since usually any model is misspecified due to hidden fac
tor(s)\, then both the proposed and the ML estimator are equally good or e
qually bad! The proposed approximations should also have applications outs
ide finance where logit type models are used and probabilities of interest
are small.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?conf
Id=5765
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5765
END:VEVENT
BEGIN:VEVENT
SUMMARY:Packing\, Combinatorial Macbeath Regions and Semi-algebraic Set Sy
stems
DTSTART;VALUE=DATE-TIME:20170726T103000Z
DTEND;VALUE=DATE-TIME:20170726T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5759@cern.ch
DESCRIPTION:The packing lemma of Haussler (J. of Comb. Theory\, Ser. A\, 1
995) states that given a set system with bounded VC dimension\, if every p
air of sets in the set system have large symmetric difference\, then the s
et system cannot contain too many sets. This has turned out to be the tech
nical foundation for many results in combinatorial geometry and discrepanc
y. Recently it was generalized by Dutta et al. (Disc. & Compt. Geom.\, 201
6) and Mustafa (Disc. & Compt. Geom.\, 2016) to the shallow packing lemma\
, applying to set systems as a function of their shallow-cell complexity.
We proved the following new results related to packings: \n\n* An optimal
lower bound for shallow packings\, thus settling the open question in Ezra
(SIAM J. on Computing\, 2016) and Dutta et al. (Disc. & Comput. Geom.\, 2
016). \n\n* Improved bounds on combinatorial Macbeath regions\, or Mnets\,
providing a combinatorial analogue to Macbeath regions in convex geometry
(Ann. of Math.\, 1952).\n\nThis resolves the main open problem in Mustafa
and Ray (Disc & Comput. Geom.\, 2016).\n\n* Mnets provide a general\, mor
e powerful framework from which the state-of-the-art unweighted Epsilon-ne
t results of Varadarajan (STOC\, 2010) and Chan et al. (SODA\, 2012) follo
w immediately.\n\n* Our upper bounds on Mnets for general semialgebraic se
t systems are asymptotically tight. We also give a more precise lower boun
d in terms of shallow-cell complexity.\n\n* Simplifying and generalizing o
ne of the main technical tools in Fox et al. (J. of the Euro. Math. Soc.\,
to appear). Besides using the packing lemma and a combinatorial construct
ion\, our proofs combine tools from polynomial partitioning of Guth and Ka
tz (Ann. of Math.\, 2014) and the probabilistic method.\n\nhttps://indico.
tifr.res.in/indico/conferenceDisplay.py?confId=5759
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5759
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Gibbsian Approach to Load Balancing in Large Graphs
DTSTART;VALUE=DATE-TIME:20170727T053000Z
DTEND;VALUE=DATE-TIME:20170727T063000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5766@cern.ch
DESCRIPTION:The talk will be on load balancing on a large graph. A unit of
load on each edge of a graph is to be distributed between its nodes in a
balanced way. On infinite graphs\, it is known that the problem exhibits n
onuniqueness. Recently\, Anantharam and Salez extended the definition of b
alanced allocations on finite graphs to their local weak limits by exploit
ing the notion of unimodularity. They used this to settle a conjecture of
Hajek on the Erdos-Renyi model. A more classical Gibbsian approach can als
o be used to arrive at the same result. This talk will highlight the key s
teps needed to make the classical approach work.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=5766
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5766
END:VEVENT
BEGIN:VEVENT
SUMMARY:A Lower Bound for Perceptrons that Compute Some Separation Problem
DTSTART;VALUE=DATE-TIME:20170728T114500Z
DTEND;VALUE=DATE-TIME:20170728T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5779@cern.ch
DESCRIPTION:We look at the boolean function in which the input is a boolea
n matrix with the promise that either 2/3rd of its rows contain a 1 or 2/3
rd of its rows do not contain a 1. The output is 1 if the matrix is of the
former kind and 0 if it is of the latter kind.\n\nAs a model of computati
on\, we look at polynomials with integer coefficients which accept an inpu
t if the polynomial evaluates to at least 1\, and rejects if the polynomia
l evaluates to at most 0. (A more rad term for such a model of computation
is a 'perceptron') Perceptrons of degree d are closely related to query a
lgorithms with at most d queries.\n\nWe prove that any perceptron computin
g the function either has large degree (at least n^{1/3}) or large coeffic
ients (at least 2^n^{1/4}).\n\nThe proof I will present is adapted from Ni
kolai Vereshchagin's original proof of the result in his paper "Lower Boun
ds for Perceptrons Solving some Separation Problems and Oracle Separation
of AM from PP". The adaptation is inspired by a proof of Thomas Watson in
his paper "Quadratic Simulations of Arthur-Merlin Games".\n\nhttps://indic
o.tifr.res.in/indico/conferenceDisplay.py?confId=5779
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5779
END:VEVENT
BEGIN:VEVENT
SUMMARY:Communication Complexity and Characterization Results in Secure Co
mputation
DTSTART;VALUE=DATE-TIME:20170731T103000Z
DTEND;VALUE=DATE-TIME:20170731T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5774@cern.ch
DESCRIPTION:Information theoretically secure multi-party computation (MPC)
has been a central primitive of modern cryptography\, in which mutually d
istrusting parties collaborate to compute a function of their private data
without revealing any additional information about their data to the othe
r parties. It is known that information theoretically secure MPC is possib
le among $n$ parties against the collusion of less than $n/2$ parties in t
he honest-but-curious model\; the threshold is $n/3$ in the malicious mode
l.\n\nDespite the huge success in many frontiers of cryptography\, many ba
sic questions in the study of secure MPC have seen a little progress. Two
of these questions are the subject of this thesis: communication and rando
mness complexity of secure MPC\; and characterization of which functions a
re securely computable. We study these questions in two and three party se
ttings against the corruption of a single party in the honest-but-curious
model. In the three party setting\, we develop new information theoretic t
ools to obtain tight lower bounds on communication and randomness complexi
ty of secure MPC. We derive lower bounds for both the perfect security cas
e (i.e.\, zero-error and no leakage of information) and asymptotic securit
y (where the probability of error and information leakage vanish as block-
length goes to infinity)\; and they are shown to be tight for several inte
resting functions.\n\nIn the two party setting\, we consider secure comput
ation of randomized functions. When both the parties (Alice and Bob) are r
equired to produce outputs\, even the characterization of which {\\em rand
omized} functions are securely computable is not known\; however\, the {\\
em deterministic} counterpart of this problem was resolved by Kushilevitz
(FOCS 1989) around three decades ago. We make some progress in this long s
tanding open problem and give a couple of characterizations: for randomize
d functions that have up to ternary output alphabet\, and for randomized f
unctions that are securely computable by up to two-round protocols. The pr
oblem becomes easier when only one party is required to produce the output
. We study this problem further in four cases: (i) when privacy is require
d against both the parties\; (ii) when privacy is required only against Al
ice\; (iii) when privacy is required only against Bob\; and (iv) when ther
e is no privacy requirement. For all the four cases we obtain optimal rate
expressions in both perfect and asymptotic security settings.\n\nhttps://
indico.tifr.res.in/indico/conferenceDisplay.py?confId=5774
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5774
END:VEVENT
BEGIN:VEVENT
SUMMARY:Derandomizing Isolation Lemma: A Geometric Approach
DTSTART;VALUE=DATE-TIME:20170801T103000Z
DTEND;VALUE=DATE-TIME:20170801T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5757@cern.ch
DESCRIPTION:We present a geometric approach towards derandomizing the Isol
ation lemma for a given family\, i.e.\, deterministically constructing a w
eight assingnment which ensures a unique minimum weight set in the family.
The idea is to work with a polytope corresponding to the family of sets.
In this talk\, we present the approach in terms of general polytopes and d
escribe a sufficient condition on the polytope for this approach to work.
The approach gives a quasi-polynomially bounded weight assignment. Finally
\, we show that two specific families - perfect matchings in bipartite gra
phs and common base sets of two matroids - satisfy the required condition
and thus\, we get an isolating weight assignment for these cases. This als
o puts the two problems in quasi-NC (based on joint works with Stephen Fen
ner and Thomas Thierauf).\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=5757
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5757
END:VEVENT
BEGIN:VEVENT
SUMMARY:General Strong-Polarization
DTSTART;VALUE=DATE-TIME:20170809T090000Z
DTEND;VALUE=DATE-TIME:20170809T100000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5780@cern.ch
DESCRIPTION:A martingale is a sequence of random variables that maintain t
heir future expected value conditioned on the past. A $[0\,1]$-bounded ma
rtingale is said to polarize if it converges in the limit to either $0$ or
$1$ with probability $1$. A martingale is said to polarize strongly\, if
in $t$ steps it is sub-exponentially close to its limit with all but expo
nentially small probability. In 2008\, Arikan built a powerful class of er
ror-correcting codes called Polar codes. The essence of his theory associa
tes a martingale with every invertible square matrix over a field (and a c
hannel) and showed that polarization of the martingale leads to a construc
tion of codes that converge to Shannon capacity. In 2013\, Guruswami and X
ia\, and independently Hassani et al. showed that strong polarization of
the Arikan martingale leads to codes that converge to Shannon capacity at
finite block lengths\, specifically at lengths that are inverse polynomial
in the gap to capacity\, thereby resolving a major mathematical challenge
associated with the attainment of Shannon capacity.\n\nWe show that a sim
ple necessary condition for an invertible matrix to polarize over any non-
trivial channel is also sufficient for strong polarization over all symmet
ric channels over all prime fields.\nPreviously the only matrix which was
known to polarize strongly was the $2\\times 2$ Hadamard matrix. In additi
on to the generality of our result\, it also leads to arguably simpler pro
ofs. The essence of our proof is a ``local definition'' of polarization wh
ich only restricts the evolution of the martingale in a single step\, and
a general theorem showing the local polarization suffices for strong polar
ization.\n\nIn this talk I will introduce polarization and polar codes and
\, time permitting\, present a full proof of our main theorem. No prior ba
ckground on polar codes will be assumed (ased on joint work with Jaroslaw
Blasiok\, Venkatesan Guruswami\, Preetum Nakkiran and Atri Rudra).\n\nhttp
s://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5780
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5780
END:VEVENT
BEGIN:VEVENT
SUMMARY:Method of Containers
DTSTART;VALUE=DATE-TIME:20170811T114500Z
DTEND;VALUE=DATE-TIME:20170811T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5812@cern.ch
DESCRIPTION:A recent powerful method has been independently developed by S
axton and Thomason (2012) and by Balogh\, Morris\, Samotij (2014). This me
thod supplies a structural characterisation of the independent sets in uni
form hypergraphs satisfying certain natural conditions\, by showing that i
n such hypergraphs every independent set is almost fully contained in one
of a small number of sparse sets (called containers). We shall see an appl
ication of the above method by Saxton and Thomason to prove the following
result (Alon'00): list chromatic number of any graph with average degree $
d$ is $\\Omega(\\log d)$.\n\nhttps://indico.tifr.res.in/indico/conferenceD
isplay.py?confId=5812
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5812
END:VEVENT
BEGIN:VEVENT
SUMMARY:Optimal Signaling Mechanisms in Unobservable Queues
DTSTART;VALUE=DATE-TIME:20170811T103000Z
DTEND;VALUE=DATE-TIME:20170811T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5813@cern.ch
DESCRIPTION:We consider the problem of optimal information sharing in the
context of a service system. In particular\, we consider an unobservable s
ingle server queue offering service at a fixed price to a Poisson arrival
of delay-sensitive customers. The service provider can observe the queue\,
and may share information about the state of the queue with each arriving
customer. The customers are Bayesian and strategic\, and incorporate any
information provided by the service provider into their beliefs about the
queue size before deciding whether to join the queue or leave without obta
ining service. We pose the following question: which signaling mechanism s
hould the service provider adopt to maximize her revenue? We establish th
at\, in general\, the optimal signaling mechanism requires the service pro
vider to strategically conceal information from the customers to incentivi
ze them to join. In particular\, we show that a signaling mechanism with t
wo signals and a threshold structure is optimal. Furthermore\, for the cas
e of linear waiting costs\, we obtain analytical expressions for the thres
holds of the optimal signaling mechanism. Finally\, we prove that the opti
mal signaling mechanism under the optimal fixed price can achieve the reve
nue of the optimal state-dependent pricing mechanism. This suggests that i
n settings where state-dependent pricing is not feasible\, the service pro
vider can effectively use optimal signaling to achieve the optimal revenue
. Our work contributes to the literature on Bayesian persuasion in dynamic
settings\, and provides many interesting directions for extensions (joint
work with David Lingenbrink\, Cornell University)).\n\nhttps://indico.tif
r.res.in/indico/conferenceDisplay.py?confId=5813
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5813
END:VEVENT
BEGIN:VEVENT
SUMMARY:Quantum Communication Using Coherent Rejection Sampling
DTSTART;VALUE=DATE-TIME:20170814T103000Z
DTEND;VALUE=DATE-TIME:20170814T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5775@cern.ch
DESCRIPTION:Compression of a message up to the information it carries is k
ey to many tasks involved in classical and quantum information theory. Sch
umacher provided one of the first quantum compression schemes and several
more general schemes have been developed ever since. However\, the one-sho
t characterization of these quantum tasks is still under development\, and
often lacks a direct connection with analogous classical tasks. Here we s
how a new technique for the compression of quantum messages with the aid o
f entanglement. We devise a new tool that we call the convex split lemma\,
which is a coherent quantum analogue of the widely used rejection samplin
g procedure in classical communication protocols. As a consequence\, we ex
hibit new explicit protocols with tight communication cost for quantum sta
te merging\, quantum state splitting. We also present a port-based telepor
tation scheme which uses less number of ports in presence of information a
bout input.\n\nVery recently our framework has found applications in sever
al important settings in quantum network theory\, such as a quantum versio
n of the Gel’fand-Pinsker channel\; the quantum broadcast channel and to
obtain a new achievability bound on quantum state redistribution\, in ter
ms of smooth-max information and hypothesis testing relative entropy. Conv
ex-split lemma has also found applications in the context of catalytic dec
oupling\; privacy in quantum communication (the wiretap channel)\; a gener
alized quantum Slepian-Wolf result \; a bound for the important and conseq
uential task of measurement compression using classical shared randomness
and to obtain optimal bounds on the classical capacity of entanglement-ass
isted compound channels.\n\nGiven the broad applicability of the convex-sp
lit technique as exhibited in these recent works\, we expect more applicat
ions in quantum network theory in the future.\n\nPaper at : http://www.com
p.nus.edu.sg/~rahul/allfiles/Coherent_rejection_sampling.pdf\n\nhttps://in
dico.tifr.res.in/indico/conferenceDisplay.py?confId=5775
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5775
END:VEVENT
BEGIN:VEVENT
SUMMARY:Using Bispectrum Invariants to Classify Translated and Rotated Ima
ges
DTSTART;VALUE=DATE-TIME:20170818T114500Z
DTEND;VALUE=DATE-TIME:20170818T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5826@cern.ch
DESCRIPTION:In this talk\, we will study the PhD-work of Risi Kondor i
n which he demonstrates how bispectrum invariants can be used to
classify translated and rotated images. The basic idea is to map the ima
ge in some 2-dimensional space to a unit sphere. Then a translation or a
rotation of the input image is given by an element of the group $SO(3)$
acting on the (image wrapped about the) sphere. Kondor shows that the b
ispectrum of the image yields a translation and rotation invariant fe
ature vector\, which can be used to classify digits (this work was done a
t CMI under the guidance of K. V. Subrahmanyam).\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=5826
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5826
END:VEVENT
BEGIN:VEVENT
SUMMARY:Distance-preserving Subgraphs of Interval Graphs
DTSTART;VALUE=DATE-TIME:20170822T103000Z
DTEND;VALUE=DATE-TIME:20170822T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5820@cern.ch
DESCRIPTION:In this talk\, we will be introducing the topic of distance-pr
eserving subgraphs\, and presenting some of our own results on distance-pr
eserving subgraphs for interval graphs (joint work with Jaikumar).\n\nThes
e results will appear in the proceedings of the conference ESA (ALGO) 2017
.\n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5820
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5820
END:VEVENT
BEGIN:VEVENT
SUMMARY:Approximating Geometric Knapsack via L-packings
DTSTART;VALUE=DATE-TIME:20170824T103000Z
DTEND;VALUE=DATE-TIME:20170824T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5777@cern.ch
DESCRIPTION:We study the two-dimensional geometric knapsack problem (2DK)\
, a geometric variant of the classical knapsack problem. In this problem w
e are given a set of axis-aligned rectangular items\, each one with an ass
ociated profit\, and an axis-aligned square knapsack. The goal is to find
a (non-overlapping) packing of a maximum profit subset of items inside the
knapsack without rotating items. This is a very well-studied optimization
problem and finds applications in scheduling\, memory allocation\, advert
isement placement etc. The best-known polynomial-time approximation factor
for this problem (even just in the cardinality case) is $2+\\eps$ [Jansen
and Zhang\, SODA 2004].\n\nAfter more than a decade\, in this paper we br
eak the 2-approximation barrier\, achieving a polynomial-time $17/9+\\eps\
n\nhttps://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5777
LOCATION: A-201 STCS Seminar Room
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5777
END:VEVENT
BEGIN:VEVENT
SUMMARY:On Shannon's Zero Error Capacity of a Graph
DTSTART;VALUE=DATE-TIME:20170901T114500Z
DTEND;VALUE=DATE-TIME:20170901T124500Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5846@cern.ch
DESCRIPTION:Suppose we want to transmit messages across a noisy channel to
a receiver. The maximum rate of transmission such that the receiver may r
ecover the original message without errors (i.e.\, zero error) is called z
ero error capacity of the channel. In this context\, a channel can be repr
esented by a graph and Shannon(1956) computed the capacity of all graphs w
ith five or fewer vertices - with the single exception of C_5 (a cycle wit
h 5 vertices). Later\, Laszlo Lovasz (1979) solved this seemingly very dif
ficult combinatorial problem by showing that capacity of C_5 is \\sqrt(5)
with an astonishingly simple argument. In this talk\, we will first develo
p the required background and discuss his proof.\n\nhttps://indico.tifr.re
s.in/indico/conferenceDisplay.py?confId=5846
LOCATION: A-201 (STCS Seminar Room)
URL:https://indico.tifr.res.in/indico/conferenceDisplay.py?confId=5846
END:VEVENT
BEGIN:VEVENT
SUMMARY:Popular Matchings in the Stable Marriage Problem
DTSTART;VALUE=DATE-TIME:20170905T103000Z
DTEND;VALUE=DATE-TIME:20170905T113000Z
DTSTAMP;VALUE=DATE-TIME:20240914T214427Z
UID:indico-event-5848@cern.ch
DESCRIPTION:The stable marriage proble