GSAS: Iterated Schr\"{o}dinger Bridge Approximation to Wasserstein Gradient Flows  

Garrett Mulcahy, University of Washington
-
PDL C-401

This talk will present a novel discretization scheme for Wasserstein gradient flows (i.e. curves of steepest descent for functions of probability measures) based on the successive computation of Schr\"{o}dinger bridges with equal marginals. Little background is assumed: just a little bit of measure theory. This talk will begin with a leisurely introduction to entropic regularized optimal transport, introducing all the notions essential for understanding this scheme. Based on joint work with Soumik Pal, Medha Agarwal, and Zaid Harchaoui.