Presentation 1999/12/11
On Successive Learning Type Algorithm for Linear Discriminant Analysis : Proposal of Learning Algorithm and Proof of Local Convergence
Kazuyuki HIRAOKA, Masashi HAMAHIRA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Linear Discriminant Analysis (LDA) is widely used in various areas, e.g. image recognition. However, when we try to apply LDA for large dimensional data, we are confronted with a problem that we have to treat N × N huge matrix when the dimension of data is N. In the present study, an online algorithm for LDA is proposed and its local convergence is proved. This algorithm has the ability of successive learning in the sense that correction of the solution according to new additional data can be executed with small computational cost. Thus it has the ability of adaptation to new environment. Moreover, our algorithm has the advantage that there is no need to maintain or calculate a huge matrix.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) linear discriminant analysis / online learning / stochastic approximation / matrix dynamics / local convergence
Paper # NC99-73
Date of Issue

Conference Information
Committee NC
Conference Date 1999/12/11(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) On Successive Learning Type Algorithm for Linear Discriminant Analysis : Proposal of Learning Algorithm and Proof of Local Convergence
Sub Title (in English)
Keyword(1) linear discriminant analysis
Keyword(2) online learning
Keyword(3) stochastic approximation
Keyword(4) matrix dynamics
Keyword(5) local convergence
1st Author's Name Kazuyuki HIRAOKA
1st Author's Affiliation Dept. of Information and Computer Sciences Saitama University()
2nd Author's Name Masashi HAMAHIRA
2nd Author's Affiliation Dept. of Information and Computer Sciences Saitama University
Date 1999/12/11
Paper # NC99-73
Volume (vol) vol.99
Number (no) 494
Page pp.pp.-
#Pages 8
Date of Issue