NeurIPS 2020

MetaPerturb: Transferable Regularizer for Heterogeneous Tasks and Architectures


Meta Review

This paper introduces a new approach for meta-learning a regularizer that empirically transfers well across different CNN architectures and image datasets. The reviewers agreed that this paper makes a worthy contribution to the NeurIPS community. The authors are encouraged to include the clarifications and new information from the author response in the camera-ready paper, including the clarification on the differences w.r.t. meta-dropout.