Deep Generative Models for Bayesian Inference in Astrophysics

Explore an in-depth astrophysics seminar where Biwei Dai from the Institute for Advanced Study discusses the application of deep generative models to Bayesian inference problems in astrophysics. The talk begins by examining how generative models can construct likelihood functions for cosmological inference at the field level, offering advantages over traditional two-point statistics, including enhanced anomaly detection and interpretability through sample generation. Learn about practical applications in weak gravitational lensing analysis and the Hyper Suprime-Cam survey. The second part demonstrates how these models create physically informed priors for Bayesian inverse problems, with a focus on reconstructing AGN accretion disk images from intensity interferometry data where phase information is missing. Discover how this approach enables high-fidelity reconstructions with uncertainty quantification that outperform traditional methods across various conditions. This hour-long seminar provides valuable insights into cutting-edge computational methods advancing astrophysical research.