NeurIPS 2020

Energy-based Out-of-distribution Detection

Meta Review

All four referees support accept. As pointed out by reviewers, AC also agree that the proposed method is easy to use and the experimental results are good (but, more architectures and non-vision datasets are encouraged to test). As also pointed out by a reviewer, JEM [10] also uses an energy-based modeling. AC also thinks that Mahalanobis [20] also uses a type of energy-based modeling. Hence, AC suggest the authors to highlight the advantage of the proposed scheme over [10] and [20] in the final draft (they did some in the rebuttal letter though). Irrespectively, AC recommend acceptance, as the proposed scheme is easier to use than [10] and [20].