Presentation 1997/11/17
Input Set Separability and Superiority of Bipolar Nets in Convergence
Hiroshi SHIRATSUCHI, Hiromu GOTANDA, Katsuhiro INOUE, Kousuke KUMAMARU,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper formulates a separability condition by normal vectors so that the separation hyperplanes should pass through a rectangle surrouding the input set. Then it depicts the distributions of two-dimensional normal vectors satisfying the condition. These distributions elucidate that the condition for the first hidden layer varies significantly even if the input patterns are simply translated, and that those conditions (in a wider sense) for the other layers are different between unipolar and bipolar units. It, also depicts an initial distribution of normal vectors with the weights initialized ordinarily by random values with zero mean. Comparison of the initial distribution to the separability condition leads to the fact that the bipolar net is superior to the unipolar net in convergence of the back propagation learning initialized randomly in such an ordinary manner. The bipolar nets exhibit better convergence than the unipolar nets even if their input space divisions in every layer by the separation hyperplanes are geometrically the same at the outset of the learning.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) multilayer nets / BP algorithm / separability of input set / convergence / bipolar nets
Paper # NC97-53
Date of Issue

Conference Information
Committee NC
Conference Date 1997/11/17(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 ENG
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Input Set Separability and Superiority of Bipolar Nets in Convergence
Sub Title (in English)
Keyword(1) multilayer nets
Keyword(2) BP algorithm
Keyword(3) separability of input set
Keyword(4) convergence
Keyword(5) bipolar nets
1st Author's Name Hiroshi SHIRATSUCHI
1st Author's Affiliation Faculty of Computer Science and Systems Engineering, Kyushu Institute of Technology()
2nd Author's Name Hiromu GOTANDA
2nd Author's Affiliation Faculty of Engineering, Kinki Univercity in Kyushu
3rd Author's Name Katsuhiro INOUE
3rd Author's Affiliation Faculty of Computer Science and Systems Engineering, Kyushu Institute of Technology
4th Author's Name Kousuke KUMAMARU
4th Author's Affiliation Faculty of Computer Science and Systems Engineering, Kyushu Institute of Technology
Date 1997/11/17
Paper # NC97-53
Volume (vol) vol.97
Number (no) 379
Page pp.pp.-
#Pages 8
Date of Issue