NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:8869
Title:Preventing Gradient Attenuation in Lipschitz Constrained Convolutional Networks


		
This paper describes a novel parametrization method for gradient-norm-preserving Lipschitz convolutional network. It is an extension of Anil et al. 2019 to the case the CNN. The paper is well-written and the technical depth qualifies NeurIPS. The proposed method seems to be a combination of exisiting ideas, so the novelty is a bit limited (but still ok for NeurIPS). The authors are encouraged to consider the reviews seriously to further improve the paper quality.