Abstract

Contributed Talk - Splinter LargeScale

Thursday, 10 September 2026, 16:30   (MW-2235)

Towards multi-tracer cosmology in wide-area surveys via scalable inference of galaxy properties

Anik Halder
University of Cambridge

Accurate redshift estimation and physically motivated sample selection are central challenges for precision cosmology with weak lensing and galaxy clustering analyses. In this talk, I will present our work on inferring physical properties of galaxies in wide-area photometric and spectroscopic surveys using a generative model for the galaxy population, pop-cosmos. This calibrated model enables principled Bayesian inference of individual galaxy redshifts and physical properties, such as stellar mass, star-formation rate, and dust content, at scale, applied to millions of sources in the Kilo-Degree Survey (KiDS-1000) and to spectroscopic galaxies in DESI. These property posteriors enable physically motivated sample splitting: in KiDS, splitting on inferred properties helps mitigate intrinsic alignment systematics in weak lensing catalogues, while in DESI it enables clustering analyses of physically distinct samples, opening a path towards multi-tracer cosmology. Overall, our results show how pop-cosmos can deliver accurate galaxy properties and sample splits at scale, connecting galaxy evolution physics to cosmological analyses with Rubin LSST and Euclid.