NeurIPS 2020

The Primal-Dual method for Learning Augmented Algorithms


Meta Review

The paper investigates a general method to augment primal-dual algorithms for incorporating predictions. The paper presents a nice theoretical and practical mix of results for a number of online optimization problems that opens possibilities to generally augment online algorithms. Reviewers agree on the rich contribution and varied content and generally good exposition. The rebuttal addressed reviewers' concerns adequately. This is a clear accept. We urge the authors to enhance the abstract to clarify what is meant by "incorporate predictions" in this context (e.g. "incorporate arbitrary solution-proposing heuristics"). The broader impact section is inadequate and should better explain how the work will help to advance ML in general.