NeurIPS 2020

Model Fusion via Optimal Transport

Meta Review

Although the reviewers had some concerns initially about baselines, small datasets, blurring the representations, etc., these were well addressed in the rebuttal and many of the reviewers have increased their scores. The consensus is that this approach will enable interesting future work on model fusion and potentially have a decent impact on areas like federated learning. Please include the requested clarifications and additional results in the final draft.