Skip to contentSkip to filters

We use cookies for analytics.

Free-Form Flows for Physics-Informed Generative Modeling

Ullrich Köthe

University of Heidelberg

Recording

51 min

Abstract

This talk first introduces a useful categorisation of the sometimes confusing "generative model zoo" in terms of different change-of-variables formulas. It then shows how normalizing flows, a major architecture for generative neural networks, can be used for simulation-based Bayesian inference in the sciences. Finally, it proposes free-form flows, which simplify the incorporation of physical prior knowledge, such as rotation and translation invariance or the restriction of a distribution to a manifold, into generative models.

Topics