The paper proposes an interesting approach and investigates the question of dataset similarity that in an important problem for transfer learning. Still it was clearly borderline in the initial review round. The feedback from the authors was appreciated and answer some of the concerns from the reviewers who reached a weak accept consensus. Note that the final version must take into account the comments from the reviewers especially the baseline comparison discussed in the feedback. More precisely the authors are expected to do the following for the final version: - Add the baselines discussed in the rebuttal in the experiments (ignoring features and labels, JDOT) and the ones discussed by R2 (distance between means) since it is an approximation of Bures and would illustrate the importance of the second order moments. - Add the discussion about JDOT and in addition to the baselines.