NeurIPS 2020

Bayesian Pseudocoresets

Meta Review

Although the reviewers consider the paper somewhat incremental, after the author response and discussion, a clear majority supports acceptance as the method appears useful and with potential to have significant impact. The reviewers note a number of weaknesses in the original submission (e.g. lack of proper reporting of privacy levels, lack of baselines for DP methods). I would strongly encourage the authors to address all of these in the final version, including ones that were omitted in the rebuttal (presumably because of lack of space).