Presentation 2000/2/3
LOGic-Oriented Back Propagation Algorithm Adjusting Decision Levels to Quantize Weight Canges
Takeshi KAMIO, Shinichiro TANAKA, Mititada MORISUE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) As both weights and neuron outputs of the logic oriented neural networks(LOGO-NNs)are quantized to integer values, it is expected that LOGO-NNs can be more effectively implemented than the common neural networks. Previosly, we have proposed the back propagation algorithm with a particular derivative for the activation function to design LOGO-NNs successfully. Although this LOGO-BP can reduce the error more rapidly than the classical BP, it is also found that this algorithm dosen't always succeed in learning. In this report, we suggest to adjust the decision levels to quantize the weight changes in order to improve LOGO-BP algorithm in the successful rate for learning. Finally, it is shown by simulations that the proposed LOGO-BP algorithm has good performance for LOGO-NNs.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) LOGO neural network / back propagation / quantized weights / multi-level threshold neuron / decision level
Paper # NC99-86
Date of Issue

Conference Information
Committee NC
Conference Date 2000/2/3(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) LOGic-Oriented Back Propagation Algorithm Adjusting Decision Levels to Quantize Weight Canges
Sub Title (in English)
Keyword(1) LOGO neural network
Keyword(2) back propagation
Keyword(3) quantized weights
Keyword(4) multi-level threshold neuron
Keyword(5) decision level
1st Author's Name Takeshi KAMIO
1st Author's Affiliation Department of Information Machines and Interfaces, Faculty of Information Sciences, Hiroshima City University()
2nd Author's Name Shinichiro TANAKA
2nd Author's Affiliation Department of Information Machines and Interfaces, Faculty of Information Sciences, Hiroshima City University
3rd Author's Name Mititada MORISUE
3rd Author's Affiliation Department of Information Machines and Interfaces, Faculty of Information Sciences, Hiroshima City University
Date 2000/2/3
Paper # NC99-86
Volume (vol) vol.99
Number (no) 612
Page pp.pp.-
#Pages 8
Date of Issue