Improved sparse low-rank matrix estimation

Ankit Parekh, Ivan Selesnick

Research output: Contribution to journalArticle

Abstract

We address the problem of estimating a sparse low-rank matrix from its noisy observation. We propose an objective function consisting of a data-fidelity term and two parameterized non-convex penalty functions. Further, we show how to set the parameters of the non-convex penalty functions, in order to ensure that the objective function is strictly convex. The proposed objective function better estimates sparse low-rank matrices than a convex method which utilizes the sum of the nuclear norm and the ℓ1 norm. We derive an algorithm (as an instance of ADMM) to solve the proposed problem, and guarantee its convergence provided the scalar augmented Lagrangian parameter is set appropriately. We demonstrate the proposed method for denoising an audio signal and an adjacency matrix representing protein interactions in the ‘Escherichia coli’ bacteria.

Original languageEnglish (US)
Pages (from-to)62-69
Number of pages8
JournalSignal Processing
Volume139
DOIs
StatePublished - Oct 1 2017

Fingerprint

Escherichia coli
Bacteria
Proteins

Keywords

  • Convex optimization
  • Low-rank matrix
  • Non-convex regularization
  • Sparse matrix
  • Speech denoising

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Software
  • Signal Processing
  • Computer Vision and Pattern Recognition
  • Electrical and Electronic Engineering

Cite this

Improved sparse low-rank matrix estimation. / Parekh, Ankit; Selesnick, Ivan.

In: Signal Processing, Vol. 139, 01.10.2017, p. 62-69.

Research output: Contribution to journalArticle

Parekh, Ankit ; Selesnick, Ivan. / Improved sparse low-rank matrix estimation. In: Signal Processing. 2017 ; Vol. 139. pp. 62-69.
@article{db5071382e60457684f3d58f7e90b89b,
title = "Improved sparse low-rank matrix estimation",
abstract = "We address the problem of estimating a sparse low-rank matrix from its noisy observation. We propose an objective function consisting of a data-fidelity term and two parameterized non-convex penalty functions. Further, we show how to set the parameters of the non-convex penalty functions, in order to ensure that the objective function is strictly convex. The proposed objective function better estimates sparse low-rank matrices than a convex method which utilizes the sum of the nuclear norm and the ℓ1 norm. We derive an algorithm (as an instance of ADMM) to solve the proposed problem, and guarantee its convergence provided the scalar augmented Lagrangian parameter is set appropriately. We demonstrate the proposed method for denoising an audio signal and an adjacency matrix representing protein interactions in the ‘Escherichia coli’ bacteria.",
keywords = "Convex optimization, Low-rank matrix, Non-convex regularization, Sparse matrix, Speech denoising",
author = "Ankit Parekh and Ivan Selesnick",
year = "2017",
month = "10",
day = "1",
doi = "10.1016/j.sigpro.2017.04.011",
language = "English (US)",
volume = "139",
pages = "62--69",
journal = "Signal Processing",
issn = "0165-1684",
publisher = "Elsevier",

}

TY - JOUR

T1 - Improved sparse low-rank matrix estimation

AU - Parekh, Ankit

AU - Selesnick, Ivan

PY - 2017/10/1

Y1 - 2017/10/1

N2 - We address the problem of estimating a sparse low-rank matrix from its noisy observation. We propose an objective function consisting of a data-fidelity term and two parameterized non-convex penalty functions. Further, we show how to set the parameters of the non-convex penalty functions, in order to ensure that the objective function is strictly convex. The proposed objective function better estimates sparse low-rank matrices than a convex method which utilizes the sum of the nuclear norm and the ℓ1 norm. We derive an algorithm (as an instance of ADMM) to solve the proposed problem, and guarantee its convergence provided the scalar augmented Lagrangian parameter is set appropriately. We demonstrate the proposed method for denoising an audio signal and an adjacency matrix representing protein interactions in the ‘Escherichia coli’ bacteria.

AB - We address the problem of estimating a sparse low-rank matrix from its noisy observation. We propose an objective function consisting of a data-fidelity term and two parameterized non-convex penalty functions. Further, we show how to set the parameters of the non-convex penalty functions, in order to ensure that the objective function is strictly convex. The proposed objective function better estimates sparse low-rank matrices than a convex method which utilizes the sum of the nuclear norm and the ℓ1 norm. We derive an algorithm (as an instance of ADMM) to solve the proposed problem, and guarantee its convergence provided the scalar augmented Lagrangian parameter is set appropriately. We demonstrate the proposed method for denoising an audio signal and an adjacency matrix representing protein interactions in the ‘Escherichia coli’ bacteria.

KW - Convex optimization

KW - Low-rank matrix

KW - Non-convex regularization

KW - Sparse matrix

KW - Speech denoising

UR - http://www.scopus.com/inward/record.url?scp=85018486108&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85018486108&partnerID=8YFLogxK

U2 - 10.1016/j.sigpro.2017.04.011

DO - 10.1016/j.sigpro.2017.04.011

M3 - Article

AN - SCOPUS:85018486108

VL - 139

SP - 62

EP - 69

JO - Signal Processing

JF - Signal Processing

SN - 0165-1684

ER -