Abstract

Much of modern cosmology can be thought of as a multi-stage optimisation problem, with the core objective of 1) constraining model parameters from data and 2) understanding where that information comes from. Simulation-based inference (SBI) utilises AI to unify simulation and parameter inference under one roof, but does not always a) leverage or b) exceed human domain knowledge in physical inference problems in terms of bits extracted from data.

I will present an information-theoretic approach to illustrate SBI , which can be naturally extended to derive hybrid statistics, an optimal framework for combining domain knowledge and learned neural summaries. These statistics improve information extraction compared to neural summaries alone or their concatenation to existing summaries and makes inference robust in settings with low training data.

We will show an application to DES Y3 weak lensing mock simulations, forecast to extract a factor of 2 more information about the dark energy equation of state than existing traditional and neural methods. We will discuss how the modular nature of hybrid statistics might shed light on where non-Gaussian signatures of Dark Energy information might lie in weak lensing maps, to be exploited in upcoming Stage IV analyses.