NeurIPS 2020

Statistical Optimal Transport posed as Learning Kernel Embedding


Meta Review

The paper addresses the problem of estimating transport map between continuous distributions through kernel mean embeddings. All reviewers agreed that the paper proposes a noveland interesting contribution to the OT literature by bridging kernel mean embeddings and OT. They achieve strong and relevant guarantees (eg dimension-free sample complexity of the transport map estimation). As such, reviewers all think that the paper has to be accepted.