Learning Theory-Informed Priors for Bayesian Inference - A Case Study with Early Dark Energy

Explore a novel method for learning theory-informed priors in Bayesian inference using normalizing flows (NF) in this 1 hour 32 minute talk by Michael Toomey from MIT. Delve into the application of this technique to early dark energy (EDE) models, which have gained attention in addressing the Hubble tension. Understand how this approach bridges the gap between theoretical cosmological models formulated in particle physics language and data analysis using physical quantities. Discover how NFs can generate priors on model parameters when analytic expressions are unavailable or complex. Learn about the validation process using limited theory-based constraints for EDE and see how this method achieves stringent constraints on EDE when incorporating large-scale structure likelihoods. Gain insights into the versatility of NFs in Bayesian inference for cosmology and beyond, and explore how generative machine learning techniques can enhance the connection between theoretical models and data analysis in physics.

Learning Theory-Informed Priors for Bayesian Inference: A Case Study with Early Dark Energy