NeurIPS 2020

Sharp Representation Theorems for ReLU Networks with Precise Dependence on Depth

Meta Review

Four knowledgeable referees rate this submission 7,7,6,6. The article presents results on the representational power of neural networks and depth separation. The referees agree that this is a topic of interest to the community and that this is a strong technical contribution albeit with limited significance to practical machine learning. They lean towards accept. I agree with the referees and hence I am recommending accept, but would like to encourage the authors to address the comments and suggestions of the referees in the preparation of the final manuscript, in particular from R1, R3, R4 about the significance.