NeurIPS 2020
### A Topological Filter for Learning with Label Noise

### Meta Review

All reviewers had a positive overall impression of this paper, and highlighted some salient features of the work:
+ generic solution to the important problem of coping with label noise
+ interesting, somewhat novel approach backed by theoretical guarantees
+ encouraging empirical results
In terms of weaknesses, it was pointed out that the empirical comparisons are only done on three datasets (CIFAR-10, 100, and MNIST), and that the discussion of Theorem 1 could be improved. The former appears reasonable for a novel idea that with a core theoretical contribution, although the authors are encouraged to incorporate elements of their response in updating the discussion around Theorem 1.