Abstract

We present a stochastic algorithm to compute the barycenter of a set ofprobability distributions under the Wasserstein metric from optimal transport.Unlike previous approaches, our method extends to continuous inputdistributions and allows the support of the barycenter to be adjusted in eachiteration. We tackle the problem without regularization, allowing us to recovera sharp output whose support is contained within the support of the truebarycenter. We give examples where our algorithm recovers a more meaningfulbarycenter than previous work. Our method is versatile and can be extended toapplications such as generating super samples from a given distribution andrecovering blue noise approximations.

 

Quick Read (beta)