Approximate joint diagonalization using a natural gradient approach

Arie Yeredor, Andreas Ziehe, Klaus Muller

Research output: Contribution to journalArticle

33 Citations (Scopus)

Abstract

We present a new algorithm for non-unitary approximate joint diagonalization (AJD), based on a "natural gradient"-type multiplicative update of the diagonalizing matrix, complemented by step-size optimization at each iteration. The advantages of the new algorithm over existing non-unitary AJD algorithms are in the ability to accommodate non-positive-definite matrices (compared to Pham's algorithm), in the low computational load per iteration (compared to Yeredor's AC-DC algorithm), and in the theoretically guaranteed convergence to a true (possibly local) minimum (compared to Ziehe et al.'s FFDiag algorithm).

Original languageEnglish
Pages (from-to)89-96
Number of pages8
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3195
Publication statusPublished - 2004 Dec 1
Externally publishedYes

Fingerprint

Diagonalization
Joints
Gradient
Iteration
Local Minima
Multiplicative
Update
Optimization

ASJC Scopus subject areas

  • Computer Science(all)
  • Biochemistry, Genetics and Molecular Biology(all)
  • Theoretical Computer Science

Cite this

Approximate joint diagonalization using a natural gradient approach. / Yeredor, Arie; Ziehe, Andreas; Muller, Klaus.

In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Vol. 3195, 01.12.2004, p. 89-96.

Research output: Contribution to journalArticle

@article{f6a090579db84acba42efb5701891aed,
title = "Approximate joint diagonalization using a natural gradient approach",
abstract = "We present a new algorithm for non-unitary approximate joint diagonalization (AJD), based on a {"}natural gradient{"}-type multiplicative update of the diagonalizing matrix, complemented by step-size optimization at each iteration. The advantages of the new algorithm over existing non-unitary AJD algorithms are in the ability to accommodate non-positive-definite matrices (compared to Pham's algorithm), in the low computational load per iteration (compared to Yeredor's AC-DC algorithm), and in the theoretically guaranteed convergence to a true (possibly local) minimum (compared to Ziehe et al.'s FFDiag algorithm).",
author = "Arie Yeredor and Andreas Ziehe and Klaus Muller",
year = "2004",
month = "12",
day = "1",
language = "English",
volume = "3195",
pages = "89--96",
journal = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
issn = "0302-9743",
publisher = "Springer Verlag",

}

TY - JOUR

T1 - Approximate joint diagonalization using a natural gradient approach

AU - Yeredor, Arie

AU - Ziehe, Andreas

AU - Muller, Klaus

PY - 2004/12/1

Y1 - 2004/12/1

N2 - We present a new algorithm for non-unitary approximate joint diagonalization (AJD), based on a "natural gradient"-type multiplicative update of the diagonalizing matrix, complemented by step-size optimization at each iteration. The advantages of the new algorithm over existing non-unitary AJD algorithms are in the ability to accommodate non-positive-definite matrices (compared to Pham's algorithm), in the low computational load per iteration (compared to Yeredor's AC-DC algorithm), and in the theoretically guaranteed convergence to a true (possibly local) minimum (compared to Ziehe et al.'s FFDiag algorithm).

AB - We present a new algorithm for non-unitary approximate joint diagonalization (AJD), based on a "natural gradient"-type multiplicative update of the diagonalizing matrix, complemented by step-size optimization at each iteration. The advantages of the new algorithm over existing non-unitary AJD algorithms are in the ability to accommodate non-positive-definite matrices (compared to Pham's algorithm), in the low computational load per iteration (compared to Yeredor's AC-DC algorithm), and in the theoretically guaranteed convergence to a true (possibly local) minimum (compared to Ziehe et al.'s FFDiag algorithm).

UR - http://www.scopus.com/inward/record.url?scp=35048855087&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=35048855087&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:35048855087

VL - 3195

SP - 89

EP - 96

JO - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

JF - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SN - 0302-9743

ER -