Presentation 1998/10/24
Energy Minimum Condition of States in Quantized Hopfield Networks for Integer Programming
Satoshi Matsuda,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Quantized Hopfield networks, where each neuron takes quantized values (e.g. integers), can obtain good solutions to integer optimization problems more quickly than the binary or continuous networks. We have, however, no theoretical justification on the relationship between the energy minimum states (solutions or nonsolutions to problems), which networks converge to, and the values of network coefficients, which needs fine-tuning. In this paper we first clarify the relationship between the dynamics of the binary and quantized networks, and then, by taking Hitchcock problems as examples of integer optimization, we theoretically show the energy minimum and nonminimum conditions of the soultions and nonslutions to the problems in terms of the values of network coefficients. This gives the insight to the tuning of network coefficients.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Hopfield network / quantized neuron / integer programming / energy minimum condition of state
Paper # NC98-51
Date of Issue

Conference Information
Committee NC
Conference Date 1998/10/24(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) Energy Minimum Condition of States in Quantized Hopfield Networks for Integer Programming
Sub Title (in English)
Keyword(1) Hopfield network
Keyword(2) quantized neuron
Keyword(3) integer programming
Keyword(4) energy minimum condition of state
1st Author's Name Satoshi Matsuda
1st Author's Affiliation Computer and Communication Research Center, Tokyo Electric Power Company()
Date 1998/10/24
Paper # NC98-51
Volume (vol) vol.98
Number (no) 365
Page pp.pp.-
#Pages 8
Date of Issue