Talk: Next-Generation Simulation-Based Inference for Robust Inference with Noisy Astrophysical Data
Speaker: Arya Farahi (Assistant Professor of Statistics and Data Science)
December 2nd
1pm-2pm CT (2-3PM ET)
Link to Zoom
Meeting ID: 886 3021 8251 / Passcode: 684859
Location: Zoom & In Person at UT Austin at Peter O'Donnell Building (POB) 4.304
Abstract: The next generation of astronomical surveys and cosmological simulations is transforming our ability to study the Universe, while introducing new challenges in extracting reliable physical information from increasingly complex, high-dimensional, and noisy observations. Traditional inference methods often rely on simplified likelihood models and assumptions that may fail to capture the complexity of astrophysical processes, observational uncertainties, and systematic effects. Simulation-based inference (SBI), combined with advances in artificial intelligence, offers a promising framework for connecting physical models to observations without requiring explicit likelihood functions.
In this talk, I will present a framework for next-generation simulation-based inference built around three complementary components. First, high-fidelity cosmological simulations, exemplified by the DREAMS project, provide physically motivated models of the Universe and enable systematic exploration of the connections between fundamental physics and observable astrophysical populations. Second, generative models learn the complex distributions encoded in these simulations, enabling efficient forward modeling of latent astrophysical populations across a broad range of physical conditions. Third, generalized Bayesian inference provides a flexible framework for comparing these synthetic populations with real observations, even when the data are noisy, incomplete, contaminated, or affected by model misspecification. Our approach enables inference through statistical comparisons between simulated and observed populations without requiring likelihood. I will highlight both the opportunities and open challenges in building AI-driven inference frameworks that are robust, scalable, and scientifically interpretable.
Speaker bio:
Arya Farahi is an Assistant Professor in the Department of Statistics and Data Science at the University of Texas at Austin, where he directs the D3 (Data, Discovery, Decision) Lab. He is leading the AI developments within Explainable Universe research initiative, and a member of the DREAMS collaboration. His research lies at the intersection of AI, statistical inference, and astrophysics, with a focus on developing robust, interpretable, and trustworthy AI methods for scientific discovery. Before joining UT Austin, he was a Data Science Fellow at the University of Michigan and a McWilliams Postdoctoral Fellow at Carnegie Mellon University. He earned his Ph.D. in Physics and Scientific Computing from the University of Michigan.