Title: Low-Rank Matrix Completion with Adversarial Missing Entries

Abstract: We give an algorithm for the adversarial matrix completion problem, in which we wish to recover a low rank matrix when adversarially chosen entries are hidden. We show that, so long as the number of missing entries in any row or column is bounded by a function of the dimension, rank, and incoherence of the matrix, nuclear norm minimization recovers the target matrix exactly. The range for which this guarantee holds is surprisingly large--in an $n\times n$ matrix of constant rank, there may be as many as $\Omega(n)$ entries missing in every row and column. Conversely, if only constantly many entries are missing in any row or column, then we may recover matrices of rank $\Omega(n)$.
As an application of adversarial matrix completion, we also give an algorithm for the problem of learning mixtures of (linearly independent) product distributions over the hypercube. If the centers of the product distribution are linearly independent, then we recover distributions with as many as $\Omega(n)$ centers in polynomial time and sample complexity.