Review for NeurIPS paper: Statistical Optimal Transport posed as Learning Kernel Embedding

Summary and Contributions: Rebuttal read. It could be good indeed to put the complexity matters in the main paper. The proposed methods consists in re-writing the original optimal transport formulation in terms of kernel embedding, such that the cost function is exactly represented (or arbitrarily close to the true one). The obtained problem then benefits from kernel framework. To estimate the marginal kernel mean embedding, the author propose a regularization based on MMD distances.