Robust reconstruction with nonconvex subset constraints: A polynomial optimization approach

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In this paper, we are interested in the recovery of an unknown signal corrupted by a linear operator, a nonlinear function, and an additive Gaussian noise. In addition, some of the observations contain outliers. Many robust data fit functions which alleviate sensitivity to outliers can be expressed as piecewise rational functions. Based on this fact, we reformulate the robust inverse problem as a rational optimization problem. The considered framework allows us to incorporate nonconvex constraints such as unions of subsets. The rational problem is then solved using recent optimization techniques which offer guarantees for global optimality. Finally, experimental results illustrate the validity of the recovered global solutions and the good quality of the reconstructed signals despite the presence of outliers.

Original languageEnglish
Title of host publicationProceedings of the 2020 IEEE 30th International Workshop on Machine Learning for Signal Processing, MLSP 2020
PublisherIEEE Computer Society
ISBN (Electronic)9781728166629
DOIs
Publication statusPublished - 1 Sept 2020
Event30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020 - Virtual, Espoo, Finland
Duration: 21 Sept 202024 Sept 2020

Publication series

NameIEEE International Workshop on Machine Learning for Signal Processing, MLSP
Volume2020-September
ISSN (Print)2161-0363
ISSN (Electronic)2161-0371

Conference

Conference30th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2020
Country/TerritoryFinland
CityVirtual, Espoo
Period21/09/2024/09/20

Keywords

  • Global optimization
  • Nonconvex constraints
  • Polynomial optimization
  • Robust estimation
  • Union of subspaces/subsets

Fingerprint

Dive into the research topics of 'Robust reconstruction with nonconvex subset constraints: A polynomial optimization approach'. Together they form a unique fingerprint.

Cite this