NeurIPS 2020

Matrix Completion with Quantified Uncertainty through Low Rank Gaussian Copula

Meta Review

This paper proposed low-rank Gaussian copula for matrix completion which enables uncertainty quantification. The reviews are divided after author response and rounds of discussions. The positive reviews focus on the theoretical characterization of confidence intervals for imputation and the scalability of the proposed model compared to other similar approaches. On the other hand, a handful of reviews pointed out the novelty of the approach is limited and it is not completely agreed upon whether this paper focuses on matrix imputation or matrix completion, which has different theoretical characterization. Another common negative point is that the related work is not thoroughly discussed. Overall, I think the pros slightly overweight the cons, especially if the authors can improve the discussion around the related work in the revision. However, I wouldn't mind if the decision is overturned.