Stochastic wasserstein barycenters

Abstract

© 2018 35th International Conference on Machine Learning, ICML 2018. All rights reserved. Wi present a stochastic algorithm to compute the baryccntcr of a set of probability distributions under the Wasscrstcin metric from optimal transport Unlike previous approaches,our method extends to continuous input distributions and allows the support of the baryccntcr to be adjusted in each iteration. VVc tacklc the problem without rcgu- larization, allowing us to rccovcr a much sharper output; We give examples where our algorithm recovers a more meaningful baryccntcr than previous work. Our method is versatile and can be extended to applications such as generating super samples from a given distribution and recovering blue noise approximations.