Convergence rates of inertial splitting schemes for nonconvex composite optimization

Patrick R. Johnstone, Pierre Moulin

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We study the convergence properties of a general inertial first-order proximal splitting algorithm for solving nonconvex nonsmooth optimization problems. Using the Kurdyka-Łojaziewicz (KL) inequality we establish new convergence rates which apply to several inertial algorithms in the literature. Our basic assumption is that the objective function is semialgebraic, which lends our results broad applicability in the fields of signal processing and machine learning. The convergence rates depend on the exponent of the 'desingularizing function' arising in the KL inequality. Depending on this exponent, convergence may be finite, linear, or sublinear and of the form O(k-p) for p > 1.

Original languageEnglish (US)
Title of host publication2017 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2017 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages4716-4720
Number of pages5
ISBN (Electronic)9781509041176
DOIs
StatePublished - Jun 16 2017
Event2017 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2017 - New Orleans, United States
Duration: Mar 5 2017Mar 9 2017

Publication series

NameICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
ISSN (Print)1520-6149

Other

Other2017 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2017
Country/TerritoryUnited States
CityNew Orleans
Period3/5/173/9/17

Keywords

  • Inertial forward-backward splitting
  • Kurdyka-Łojaziewicz Inequality
  • convergence rate
  • first-order methods
  • heavy-ball method

ASJC Scopus subject areas

  • Software
  • Signal Processing
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Convergence rates of inertial splitting schemes for nonconvex composite optimization'. Together they form a unique fingerprint.

Cite this