NeurIPS 2020

Locally-Adaptive Nonparametric Online Learning

Meta Review

There were four reviews, of which one was initially proposing rejecting and three accepting. The submission addresses an important direction for the development of online algorithms, uses an interesting mix of techniques and achieves good bounds, and the setting is fairly general. Some concerns of the reviewers, such as those regarding running time and hyperparameters, were answered by the authors to the reviewers' satisfaction. However, an issue that came up in discussing the authors' reply was the space complexity, which is O(T). While this is not too bad (even for an online algorithm) and even an inefficient algorithm would serve as a basis for further development, this is still a somewhat negative factor in the decision. In the end, the reviewers reached a consensus for accepting the paper.