Presentation 1999/11/19
A method to select mainly misclassified patterns in Learning Vector Quantization
Takahiko SHIMMURA, Hiroshi IZUMOTO, Hitoshi / SAKANO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, we propose a method to select mainly misclassified patterns in Learning Vector Quantization (LVQ). The method to select patterns with the Apollonian hyperspheres has been proposed because it is effective to learn mainly patterns near a discriminating plane by LVQ. This method virtually brings about sluggish learning because it arbitrarily chooses patterns. We propose the above method as a remedy for the trouble. The new method has been applied to learning of a dictionary to recognize handwritten Japanese characters, and so a recognition rate by this method is higher than that by Apollonian hyperspheres. By analyzing selected patterns, we find the method to be effective in solving the trouble.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) character recognition / LVQ / selection of learning patterns
Paper # PRMU99-145
Date of Issue

Conference Information
Committee PRMU
Conference Date 1999/11/19(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 Pattern Recognition and Media Understanding (PRMU)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A method to select mainly misclassified patterns in Learning Vector Quantization
Sub Title (in English)
Keyword(1) character recognition
Keyword(2) LVQ
Keyword(3) selection of learning patterns
1st Author's Name Takahiko SHIMMURA
1st Author's Affiliation NTT Data Corporation()
2nd Author's Name Hiroshi IZUMOTO
2nd Author's Affiliation NTT Data Corporation
3rd Author's Name Hitoshi / SAKANO
3rd Author's Affiliation NTT Data Corporation /
Date 1999/11/19
Paper # PRMU99-145
Volume (vol) vol.99
Number (no) 449
Page pp.pp.-
#Pages 6
Date of Issue