NeurIPS 2020

Lipschitz-Certifiable Training with a Tight Outer Bound

Meta Review

The paper proposes a certified robust training method based on Lipschitz bounds. The reviewers initially have concerns due to lack of clarity and problematic experimental comparisons. Those concerns were addressed by the rebuttal and all the reviewers agreed to weakly accept the paper since it outperforms existing methods on L2 certified robust training. We hope the authors can improve clarity and carefully incorporate the rebuttal into the main paper.