NeurIPS 2020

Calibrating Deep Neural Networks using Focal Loss

Meta Review

This paper was reviewed by 4 reviewers and there was unanimous agreement that the paper should be accepted (scores ranging from "marginally above threshold" to "clear accept"). I agree with the reviewers and recommend the paper be accepted. All 4 reviewers provided quite detailed suggestions for improvement in the paper and I strongly recommend that the authors carefully take these suggestions into account in revising the paper for the camera-ready version. In particular, please be sure to please take into account the reviewer suggestions (R1 and R2 specifically) on improving and clarifying the wording of your OOD claims. And please also be sure to include vanilla Focal Loss results in the main paper (R1, R2, R5).