Motivation of the method

Neural Information Processing Systems 

Following Reviewer #3, we clarify the motivation behind NC. General changes to the manuscript Following Reviewer's #1 suggestion, we included in the Appendix our experi-5 Our intent was to reason about optimal discriminators. Following Reviewer's #1 and #3 remarks, we replace the Donsker-V aradhan lower We thank the reviewers for their careful reading. We then use Jensen's inequality with uniform We follow Reviewer's #1 suggestion to