AstroAI Workshop 2025
Philipp Frank
Probabilistic Inference in Astrophysics: Variational, Flow-Based, and Diffusion Models
Presenter: Philipp Frank (Stanford University)
Title: Probabilistic Inference in Astrophysics: Variational, Flow-Based, and Diffusion Models
Date/Time: Tuesday, July 8th, 3:30 - 5:00 PM
Abstract: Probabilistic modeling provides a principled framework for reasoning under uncertainty—a core requirement in many areas of astrophysics, where we face noisy, incomplete, and indirect observations of complex physical systems. This tutorial offers a conceptual and practical introduction to Bayesian modeling, with a focus on the key ingredients: priors, likelihoods, posteriors, and strategies for inference.
In practice, many astrophysical inference problems involve high-dimensional and structured posteriors, shaped by physical symmetries, selection effects, and latent variables. Exploring or approximating these posteriors poses both computational and methodological challenges. In this session, we will survey a range of modern tools designed to address these challenges—including variational methods, Markov chain Monte Carlo, normalizing flows, and diffusion-based generative models. Rather than prescribing a single solution, we will explore the assumptions, trade-offs, and conceptual foundations of each approach. We will see how these methods can be understood through the lens of probabilistic inference, and how they can be applied to practical problems such as simulation-based inference, posterior sampling, and generative modeling in astrophysics. Throughout, we will highlight real-world use cases and discuss how to align modeling choices with scientific goals.
Requirements: None