NeurIPS 2020

CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instances


Meta Review

The paper addresses the problem of out of distribution (OOD) detection for computer vision by making sensible changes to the contrastive learning framework. The reviewers agree that the contributions are sufficient for acceptance at NeurIPS, and I am willing to ignore the comment about novelty, given that even though the approach is simple, its effectiveness in the context of OOD detection was now well established. PS: it makes sense for this paper and arxiv.org/abs/2007.05566 to cite each other as concurrent work and discuss the similarities and differences.