English

Matrix Completion from $O(n)$ Samples in Linear Time

Machine Learning 2017-08-23 v4 Data Structures and Algorithms Machine Learning Optimization and Control

Abstract

We consider the problem of reconstructing a rank-kk n×nn \times n matrix MM from a sampling of its entries. Under a certain incoherence assumption on MM and for the case when both the rank and the condition number of MM are bounded, it was shown in \cite{CandesRecht2009, CandesTao2010, keshavan2010, Recht2011, Jain2012, Hardt2014} that MM can be recovered exactly or approximately (depending on some trade-off between accuracy and computational complexity) using O(npoly(logn))O(n \, \text{poly}(\log n)) samples in super-linear time O(napoly(logn))O(n^{a} \, \text{poly}(\log n)) for some constant a1a \geq 1. In this paper, we propose a new matrix completion algorithm using a novel sampling scheme based on a union of independent sparse random regular bipartite graphs. We show that under the same conditions w.h.p. our algorithm recovers an ϵ\epsilon-approximation of MM in terms of the Frobenius norm using O(nlog2(1/ϵ))O(n \log^2(1/\epsilon)) samples and in linear time O(nlog2(1/ϵ))O(n \log^2(1/\epsilon)). This provides the best known bounds both on the sample complexity and computational complexity for reconstructing (approximately) an unknown low-rank matrix. The novelty of our algorithm is two new steps of thresholding singular values and rescaling singular vectors in the application of the "vanilla" alternating minimization algorithm. The structure of sparse random regular graphs is used heavily for controlling the impact of these regularization steps.

Keywords

Cite

@article{arxiv.1702.02267,
  title  = {Matrix Completion from $O(n)$ Samples in Linear Time},
  author = {David Gamarnik and Quan Li and Hongyi Zhang},
  journal= {arXiv preprint arXiv:1702.02267},
  year   = {2017}
}

Comments

45 pages, 1 figure. Short version accepted for presentation at Conference on Learning Theory (COLT) 2017

R2 v1 2026-06-22T18:12:18.488Z