Bayesian cosmological analysis with full forward modeling of galaxy data-sets

Cosmological data-sets, such as galaxy surveys, are complex to use owed to the limits imposed by the instrumentation and unwanted systematic effectsinduced by our environment (e.g. fiber collisions, atmosphere, dust reddening, stars, scanning patterns). Most of the effort of the past decades has leaned on pushing the use of summary statistics (e.g. to point correlation function, galaxy cluster abundance) to do global inference on either cosmology or on the tracer properties (e.g. galaxies, QSOs). However such an operation drops a large fraction of the information of the data- Sets, and make the analysis conceptually less easy to apprehend.

Over the past decade, frameworks have risen to handle in a more direct way the interpretation of those data-sets by relying fully on Bayes identity. One of those framework, BORG, is based on the principle that we can model in detail the distribution of galaxies from statistically simple initial conditions then we can build an optimal inference machine for cosmology. I 
will show the principle on which this machine is built on, highlights successful application to data and discuss some limitations.