Translator Disclaimer
2015 Adaptive multinomial matrix completion
Olga Klopp, Jean Lafond, Éric Moulines, Joseph Salmon
Electron. J. Statist. 9(2): 2950-2975 (2015). DOI: 10.1214/15-EJS1093

Abstract

The task of estimating a matrix given a sample of observed entries is known as the matrix completion problem. Most works on matrix completion have focused on recovering an unknown real-valued low-rank matrix from a random sample of its entries. Here, we investigate the case of highly quantized observations when the measurements can take only a small number of values. These quantized outputs are generated according to a probability distribution parametrized by the unknown matrix of interest. This model corresponds, for example, to ratings in recommender systems or labels in multi-class classification. We consider a general, non-uniform, sampling scheme and give theoretical guarantees on the performance of a constrained, nuclear norm penalized maximum likelihood estimator. One important advantage of this estimator is that it does not require knowledge of the rank or an upper bound on the nuclear norm of the unknown matrix and, thus, it is adaptive. We provide lower bounds showing that our estimator is minimax optimal. An efficient algorithm based on lifted coordinate gradient descent is proposed to compute the estimator. A limited Monte-Carlo experiment, using both simulated and real data is provided to support our claims.

Citation

Download Citation

Olga Klopp. Jean Lafond. Éric Moulines. Joseph Salmon. "Adaptive multinomial matrix completion." Electron. J. Statist. 9 (2) 2950 - 2975, 2015. https://doi.org/10.1214/15-EJS1093

Information

Received: 1 July 2014; Published: 2015
First available in Project Euclid: 5 January 2016

zbMATH: 1329.62304
MathSciNet: MR3439190
Digital Object Identifier: 10.1214/15-EJS1093

Subjects:
Primary: 62J02, 62J99
Secondary: 62H12,60B20

Rights: Copyright © 2015 The Institute of Mathematical Statistics and the Bernoulli Society

JOURNAL ARTICLE
26 PAGES


SHARE
Vol.9 • No. 2 • 2015
Back to Top