A1333
Title: Empirical Bayes 1-bit matrix completion
Authors: Takeru Matsuda - University of Tokyo & RIKEN Center for Brain Science (Japan) [presenting]
Abstract: The problem of predicting unobserved entries of a binary data matrix from observed entries is called the 1-bit matrix completion. An empirical Bayes method is developed for 1-bit matrix completion that utilizes a low-rank structure like the multidimensional item response theory. The proposed method is motivated by an empirical Bayes estimator of a normal mean matrix of a prior study, which is a matrix generalization of the James--Stein estimator and shrinks the singular values towards zero. Simulation results and application to real data are presented.
