NeurIPS 2020

Efficient active learning of sparse halfspaces with arbitrary bounded noise

Meta Review

All reviewers agree that this paper made a solid contribution in the context of active learning of sparse halfspaces (in the Massart noise model). The sample complexity bound amounts to a major improvement over best known results for learning of halfspaces, which warrant acceptance of the paper. For camera-ready, the authors are encouraged to take into account the reviewer's feedback to further improve the discussion of the proposed algorithm (In particular, please address the concern on lack of intuition why a mirror descent approach leads to such large improvements in this space compared to earlier methods).