Reviews: On the Power and Limitations of Random Features for Understanding Neural Networks

This would require a new review that cannot be performed in the current conference cycle. Regarding your answer to 3b,c this notation is really misleading if you want to mean the norm of a function then you should use f not f(x) *** Original review *** Originality. The authors focus on negative results that show some fundamental limits on what can be learned with random features, which tend to deviate from previous works that rather aim to provide positive results. Understanding the limits of random features is a valuable avenue of research. However I find that the included positive result (Theorem 3.1) is difficult to assess as a significant contribution as even the authors claim that it is not fundamentally novel and is a rather new/more compact/better proof than that of other similar results. I find there are some major flaws in some parts, or either I am misunderstanding something.