NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:3568
Title:On the Power and Limitations of Random Features for Understanding Neural Networks


		
This paper shows that random feature methods can not efficiently learn even a single ReLU. This has stark implications for a lot of recent work trying to explain the success of deep learning through random feature methods. The authors however should take into account the reviews and improve the writing and presentation.