Supranta S. Boruah is a Center for Particle Cosmology fellow at the University of Pennsylvania. He works on developing statistical and machine learning methods for cosmological analysis. Supranta is interested in various aspects of large-scale structure cosmology. Currently, his main research focus is on developing field-level analysis methods for weak lensing data. He has also developed fast machine-learning based emulators for 3x2 pt analysis and have worked on extracting cosmological information from peculiar velocity data.
Title: Optimal weak lensing and galaxy clustering analysis with field-level inference
Abstract: Traditional cosmological analyses rely on summary statistics, which may lead to lossy compression, discarding valuable information encoded in the full cosmological field. Field-level inference offers a path to optimal cosmological constraints by directly modeling the complete density field, but its success hinges critically on accurate forward models and realistic priors. This talk presents our efforts to unlock field-level inference for weak lensing surveys through advances in both methodology and applications to real observational data. First, I describe KaRMMa, our code for reconstructing Bayesian weak lensing mass maps with lognormal priors that capture non-Gaussian structure formation, which we have successfully applied to Dark Energy Survey Year 3 (DES-Y3) data. Second, I discuss how we enhance KaRMMa's forward model using improved analytical prescriptions for small-scale physics. Third, I demonstrate how generative machine learning models—including GANs and diffusion models—can learn data-driven priors that surpass analytical approximations. I will show that diffusion models applied to DES-Y3 enables smaller-scale reconstructions of cosmic structures from real survey data. Finally, I outline our roadmap toward joint field-level inference combining weak lensing and galaxy clustering, with particular emphasis on modeling galaxy bias at the field level. These advances, validated on current survey data, promise to extract maximal cosmological information from upcoming surveys like Rubin Observatory and Roman Space Telescope.