New statistical and ML-based methods for weak lensing and galaxy clustering analysis for Stage-IV surveys

With the increased observational precision of Stage-IV dark energy surveys, we will face new statistical and data analysis challenges. I will talk about two such challenges for weak lensing and galaxy clustering analysis and our efforts to solve these problems. First, preparing for standard 2-point analysis requires running hundreds (maybe even thousands) of MCMC chains, which will form a substantial computational bottleneck for the Stage-IV surveys. I will present a new iterative emulator method using neural networks that leads to fast and efficient inference in high dimensional parameter spaces – thus solving a major computational problem for Stage-IV data analysis. I will also talk about some recent investigations into the impact of systematics on LSST 3x2 pt analysis. Next, 2-point analyses are necessarily suboptimal as the late time cosmic density is highly non- Gaussian. I will talk about our efforts to build a Bayesian map-level inference method to analyse weak lensing data. I will talk about the recent progress, challenges and promises of such field-level analysis which is the optimal way to extract information from a data set at a given scale.