New dynamical systems for principal and minor subspace analysis

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

Abstract

New dynamical systems for extracting multiple principal and minor components of a square matrix are presented. Analyses for determining invariant sets, domains of attraction and asymptotic stability of these systems are provided. These systems can be slightly modified so that they converge to the actual eigen or singular vectors by incorporating a diagonal matrix having distinct eigenvalues. Some of the proposed algorithms generalize known systems such as Oja's systems for principal and minor component analysis and are derived from optimizing bounded function over compact sets. Dual purpose systems for computing minor and principal component analyzers are also derived. Additionally, exact solutions for some non-linear learning dynamical systems are given

Original languageEnglish (US)
Title of host publicationProceedings of the 45th IEEE Conference on Decision and Control 2006, CDC
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages5180-5185
Number of pages6
ISBN (Print)1424401712, 9781424401710
DOIs
StatePublished - 2006
Event45th IEEE Conference on Decision and Control 2006, CDC - San Diego, CA, United States
Duration: Dec 13 2006Dec 15 2006

Publication series

NameProceedings of the IEEE Conference on Decision and Control
ISSN (Print)0743-1546
ISSN (Electronic)2576-2370

Other

Other45th IEEE Conference on Decision and Control 2006, CDC
Country/TerritoryUnited States
CitySan Diego, CA
Period12/13/0612/15/06

Keywords

  • Analytic solutions
  • Asymptotic stability
  • Dynamical system
  • Exact solutions
  • Global convergence
  • Gradient flow
  • Invariant set
  • Lasalle invariance principle
  • MCA
  • MSA
  • Minor subspace flow
  • Oja's rule
  • PCA
  • PSA
  • Principal subspace flow
  • Stiefel manifold

Fingerprint

Dive into the research topics of 'New dynamical systems for principal and minor subspace analysis'. Together they form a unique fingerprint.

Cite this