52
7

A non-backtracking method for long matrix and tensor completion

Abstract

We consider the problem of low-rank rectangular matrix completion in the regime where the matrix MM of size n×mn\times m is ``long", i.e., the aspect ratio m/nm/n diverges to infinity. Such matrices are of particular interest in the study of tensor completion, where they arise from the unfolding of a low-rank tensor. In the case where the sampling probability is dmn\frac{d}{\sqrt{mn}}, we propose a new spectral algorithm for recovering the singular values and left singular vectors of the original matrix MM based on a variant of the standard non-backtracking operator of a suitably defined bipartite weighted random graph, which we call a \textit{non-backtracking wedge operator}. When dd is above a Kesten-Stigum-type sampling threshold, our algorithm recovers a correlated version of the singular value decomposition of MM with quantifiable error bounds. This is the first result in the regime of bounded dd for weak recovery and the first for weak consistency when dd\to\infty arbitrarily slowly without any polylog factors. As an application, for low-CP-rank orthogonal kk-tensor completion, we efficiently achieve weak recovery with sample size O(nk/2)O(n^{k/2}) and weak consistency with sample size ω(nk/2)\omega(n^{k/2}). A similar result is obtained for low-multilinear-rank tensor completion with O(nk/2)O(n^{k/2}) many samples.

View on arXiv
Comments on this paper