NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:8746
Title:Momentum-Based Variance Reduction in Non-Convex SGD


		
The paper proposes a new momentum-based variance reduction algorithm for stochastic non-convex optimization. All reviewers are very positive about the paper. I am very happy to recommend acceptance as a poster and congratulate the authors on a nice piece of work. In the camera ready, please address the review comments, especially add the comparison with other related work as suggested by one reviewer. Regarding the lower bound result for nonconvex finite-sum optimization, the following recent paper might be of interest: Zhou, D., & Gu, Q. (2019). Lower bounds for smooth nonconvex finite-sum optimization. ICML. [This meta-review was reviewed and revised by the Program Chairs]