NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:1217
Title:An Improved Analysis of Training Over-parameterized Deep Neural Networks

This paper analyzes the convergence of GD and SGD for overparametrized networks, which result in improvements of the overparametrization requirement. Initially the paper received weakly positive reviews. Specifically, they felt that while the contribution follows closely prior work [2,16], the paper still makes a nice contribution, which is insightful and novel. The rebuttal addressed the issues raised by the reviewers. While one reviewer remained concerned as to whether the ideas in the paper can be extended further, upon discussion, the reviewers agreed that the paper should be accepted.