Title:
LINKAGE LEARNING DEVICE FOR ARTIFICIAL NEURAL CIRCUIT NETWORK
Document Type and Number:
Japanese Patent JP3364242
Kind Code:
B2
Abstract:
PURPOSE: To provide a device capable of learning the weight of the linkage of an artificial neural circuit network so as to properly remember even for a bias pattern in an associative memory.
CONSTITUTION: The energy for a value of a learning pattern is calculated (109) and the learning quantity of the weight of the linkage is varied (110) in accordance with the difference between it and a target value (111). But, when the difference is small, learning is omitted. Consequently, the lower the energy of the pattern is, the more easily it is remembered in the associative memory. Thus, it is difficult to remember the pattern of high energy. But the pattern can be learned so as to take equal energy for every storage pattern by the piece of constitution so that even such a bias pattern as make difference in energy can properly be remembered.
Inventors:
Suzuoka Setsu
Application Number:
JP17653692A
Publication Date:
January 08, 2003
Filing Date:
July 03, 1992
Export Citation:
Assignee:
Toshiba Corporation
International Classes:
G06F15/18; G06F9/44; G06G7/60; G06N3/04; G06N3/08; G06N5/04; G06N99/00; (IPC1-7): G06N3/08; G06N3/04
Other References:
天野真家・他，「ニューラルネットワークを用いたかな漢字変換技術」，東芝レビュー，日本，株式会社東芝，１９９０年１２月１日，Ｖｏｌ．４５，Ｎｏ．１２，ｐｐ．９８７−９９０
西山清・他，「冗長なニューロンをもつＨｏｐｆｉｅｌｄニューラルネットワークに基づく連想記憶モデル（ＩＩＩ）」，電子情報通信学会技術研究報告，日本，社団法人電子情報通信学会，１９９２年３月１９日，Ｖｏｌ．９１，Ｎｏ．５３０（ＮＣ９１−１３２〜１６４），ｐｐ．１−８
鈴岡節，「連想メモリにおけるスパースにエンコードさらたパターンの連想能力についての一考察」，電子情報通信学会技術研究報告，日本，社団法人電子情報通信学会，１９９２年１月１８日，Ｖｏｌ．９１，Ｎｏ．４１４（ＮＣ９１−８２〜９７），ｐｐ．６５−７０
西山清・他，「冗長なニューロンをもつＨｏｐｆｉｅｌｄニューラルネットワークに基づく連想記憶モデル（ＩＩＩ）」，電子情報通信学会技術研究報告，日本，社団法人電子情報通信学会，１９９２年３月１９日，Ｖｏｌ．９１，Ｎｏ．５３０（ＮＣ９１−１３２〜１６４），ｐｐ．１−８
鈴岡節，「連想メモリにおけるスパースにエンコードさらたパターンの連想能力についての一考察」，電子情報通信学会技術研究報告，日本，社団法人電子情報通信学会，１９９２年１月１８日，Ｖｏｌ．９１，Ｎｏ．４１４（ＮＣ９１−８２〜９７），ｐｐ．６５−７０
Attorney, Agent or Firm:
Hideaki Togawa