The goal of the workshop is to support focused, interdisciplinary discussions that can help unify recent advances in generative models relying on measure transport and diffusion processes in the context of sampling and MCMC. A complete abstract for the workshop is appended below.
Abstract: This workshop aims to foster discussions in machine learning
underlying modern generative models and their role in MCMC-like
sampling algorithms as well as explore how viewpoints arising in
adjacent fields can be used to refine and approach remaining open
challenges in the contexts of probabilistic modeling and
high-dimensional sampling. The past decade has seen a surge of
progress in the empirical performance of techniques such as diffusion
models and normalizing flows, but much of the success of these
techniques amounts to careful consideration of how to define a map
between distributions. This topic has a substantial history in the
fields of optimal transport, stochastic processes, and variational
inference. By bringing together theorists and practitioners from these
camps, we hope to clarify the perspectives of recent advances across
their respective communities.
Organizers:
Michael Albergo
Bob Carpenter
Neha Wadia
Nick Boffi
Joan Bruna
Measure Transport, Diffusion Processes and Sampling Event Website