NeurIPS 2020

Long-Tailed Classification by Keeping the Good and Removing the Bad Momentum Causal Effect

Meta Review

This paper proposes an interesting new perspective on imbalanced learning, focusing on the effect of momentum in SGD. The theoretical motivation is paired with a thorough ablation and empirical results. There were some weaknesses cited by the reviewers, including more thorough comparisons and some details regarding the theoretical part. The reviewers felt that the authors addressed these concerns in the rebuttal. As a result, this paper adds a strong contribution that could open up new perspectives for the problem of imbalance. The authors should make sure to include the clarifications and additional details in the camera-ready version.