AstroAI Lunch Talks - April 28, 2025 - Arne Thomsen
28 Apr 2025 - Joshua Wing
The video can be found here: https://www.youtube.com/watch?v=FX3NMp_1z_M
Speaker: Arne Thomsen (ETH Zurich)
Title: Simulation-based inference of cosmology from multi-probe maps using deep learning
Abstract: The ongoing Dark Energy Survey (DES) is designed to observe the large-scale structure of the Universe using a number of cosmological probes, including weak gravitational lensing and galaxy clustering. Conventionally, constraints on the cosmological parameters are calculated by comparing two-point functions of the observables with semi-analytical theory predictions. However, we know that due to nonlinear structure formation at late times, the physical fields contain non-Gaussian information which is not captured by the two-point functions. In this talk, I present a pipeline to leverage numerical theory predictions and the expressive power of deep learning to extract this additional cosmological information by learning the (beyond Gaussian) summary statistic instead. More specifically, we present our analysis framework for simulation-based inference in three parts: 1) a forward model to generate DES-like weak gravitational lensing and galaxy clustering maps from the CosmoGridV1 suite of simulations, 2) a pipeline to train artificial neural networks to find low-dimensional features that maximize the cosmological information gain from these forward-modeled maps, 3) a flexible inference workflow to obtain the posterior distribution of the cosmological and astrophysical parameters, given a (mock) observation. So far, we have applied the above to synthetic observations for forecasting and validation, comparing our approach to a power-spectrum baseline. We are currently preparing the whole pipeline for application to the real survey data, putting an emphasis on its validation.