Information Theory of Matrix Completion

Abstract

Matrix completion is a fundamental problem that comes up in a variety of applications like the Netflix problem, collaborative filtering, computer vision, and crowdsourcing. The goal of the problem is to recover a k-by-n unknown matrix from a subset of its noiseless (or noisy) entries. We define an information-theoretic notion of completion capacity C that quantifies the maximum number of entries that one observation of an entry can resolve. This number provides the minimum number m of entries required for reliable reconstruction: m=kn/C. Translating the problem into a distributed joint source-channel coding problem with encoder restriction, we characterize the completion capacity for a wide class of stochastic models of the unknown matrix and the observation process. Our achievability proof is inspired by that of the Slepian-Wolf theorem. For an arbitrary stochastic matrix, we derive an upper bound on the completion capacity.