A Study on Clustering Method Using the Wasserstein Barycenter

Abstract

The Wasserstein barycenter is the mean of several probability distributions based on the Wasserstein distance. When there is only one input distribution, the algorithm for finding the Wasserstein barycenter is known to be consistent with the k-means algorithm. In this study, numerical experiments were conducted to compare clustering using the Wasserstein barycenter with existing clustering methods. Numerical results show that clustering with Wasserstein barycenter is comparable to existing clustering methods in ARI.