NeurIPS 2020

Fast geometric learning with symbolic matrices

Meta Review

Reviewers agree that the central problem addressed by this paper is of value, and all reviewers recommend to accept the paper. The rebuttal makes clear that the potential application is beyond just distance calculations, and I agree with the positive reviewers that there is a potential to influence practice in areas of ML across the NeurIPS audience. I disagree that "we never intended to compete with [approx NN]" was "made this clear in the paper (Section 2, 4, 5.2, and the legend of Table 2)". I believe it was rather quite obliquely implied. There is no mention in "limitations" of the consequences of what R1 refers to as the "bruteforce" comparison. I believe that being clearer on this point would not detract at all from the paper, and indeed would strengthen it.