Gradient descent algorithms for Bures-Wasserstein barycenters

MaRDI QIDQ6332404zbMATH OpenFDO

Authors Sinho Chewi, Tyler Maunu, Philippe Rigollet, Austin J. Stromme

Publication date 6 January 2020

Mathematics Subject Classification ID

Variational problems in infinite-dimensional spaces (58Exx) Point estimation (62F10) Approximation algorithms (68W25) Nonconvex programming, global optimization (90C26)

Abstract: We study first order methods to compute the barycenter of a probability distribution $P$ over the space of probability measures with finite second moment. We develop a framework to derive global rates of convergence for both gradient descent and stochastic gradient descent despite the fact that the barycenter functional is not geodesically convex. Our analysis overcomes this technical hurdle by employing a Polyak-Lojasiewicz (PL) inequality and relies on tools from optimal transport and metric geometry. In turn, we establish a PL inequality when $P$ is supported on the Bures-Wasserstein manifold of Gaussian probability measures. It leads to the first global rates of convergence for first order methods in this context.

This page was built for publication: Gradient descent algorithms for Bures-Wasserstein barycenters

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6332404)