NeurIPS 2020

CoinPress: Practical Private Mean and Covariance Estimation

Meta Review

All four reviewers support acceptance of this paper. They agree that the paper makes both empirical and theoretical progress on the problem of mean and covariance estimation under differential privacy. I therefore recommend accept. In the camera ready, the authors should include discussion of the t parameter of their algorithm (i.e., how to set it to get good performance, and clarify that t=1 is the Analyze Gauss Baseline), mention generalizations beyond sub-Gaussian distributions, and clarify the differences to Kamath, Li, Singhal, Ullman [24].