Reviews: Wasserstein Weisfeiler-Lehman Graph Kernels

The main motivation of this work is based on the fact that conventional graph kernels loose information in their embedding and/or aggregation steps. While we agree with the authors on this point, it is not clear what is the information lost with the proposed WWL graph kernel. Since the proposed method is based on the WL subtree kernel, then it has the same weaknesses as it. Moreover, it may have more issues, such as the non-uniqueness of the embedding, the iterative operations related to hashing… The part "To ensure the theoretical correctness of our results…" is confusing and misleading. On a first reading, the reader may understand that the theoretical results are not correct.