A0202
Title: Conformalized matrix completion
Authors: Yu Gui - University of Chicago (United States) [presenting]
Rina Foygel Barber - University of Chicago (United States)
Cong Ma - University of Chicago (United States)
Abstract: Matrix completion aims to estimate missing entries in a data matrix using the assumption of a low-complexity structure (e.g., low rank) to make imputation possible. While many effective estimation algorithms exist in the literature, uncertainty quantification for this problem has proved challenging, and existing methods are extremely sensitive to model misspecification. A distribution-free method is proposed for predictive inference in the matrix completion problem. The method adapts the framework of conformal prediction, which provides confidence intervals with guaranteed distribution-free validity in the regression setting, to the matrix completion problem. The resulting method, conformalized matrix completion (CMC), offers provable predictive coverage regardless of the accuracy of the low-rank model. Empirical results on simulated and real data demonstrate that CMC is robust to model misspecification while matching the performance of existing model-based methods when the model is correct.
