Efficient Statistical Models for Generating High-Resolution Black Hole Images

Interpreting data from the Event Horizon Telescope (EHT) has significantly advanced our understanding of black holes and their surrounding plasma. Nonetheless, discrepancies remain between predicted and observed source-integrated variability in Sgr A*. In this talk, as a way to tackle this variability issue, we present a framework that combines phenomenological statistical models with generative techniques capable of efficiently generating black hole images and resolve higher-order photon rings. This framework results from training separate conditional generative adversarial networks to generate high-resolution black hole images from fluid and stochastic simulations that are then used to study temporal and spatial correlation structures of accretion flows. Given the uncertainty associated with the variability excess and the possibility of missing physical ingredients in current astrophysical models, developing these types of astrophysics-agnostic phenomenological approaches is as valuable as enhancing accretion disk simulations.