NeurIPS 2020

Optimal Iterative Sketching Methods with the Subsampled Randomized Hadamard Transform

Meta Review

Four knowledgeable reviewers recommend accept, and I concur, in light of the contributions made: a sharp asymptotic analysis of the behavior of SRHT and Haar sketches, and the provision of optimal step sizes. Please revise your paper in accordance with the promises made in the author rebuttal, including the provision of numerical results on the effects of not refreshing the sketch at each iteration.