Զѧ
ACTA AUTOMATICA SINICA
1997ꡡ236ڡVol.23No.61997




һЧѧϰ㷨
ѧ¡ῡ
ժҪһµģ͵Чѧϰ㷨ص㣺(1)ȫ洢ѵģʽѵģʽĿ֮Եǿûƣ(2)Сѵģﵽ(3)(2)Ļϣÿѵģʽоܴ(4)ȫȶ.ļʵ˵ѧϰ㷨㷨иǿĴ洢ݴ.
ؼʡ磬ѵģʽȫ洢ȫȶ.
AN EFFICIENT LEARNING ALGORITHM FOR ASSOCIATIVE
MEMORY NEURAL NETWORK
LIANG XUEBINWU LIDEYU JUN
(Dept.of Computer ScienceFudan University,Shanghai200433)
AbstractA new and efficient learning algorithm of asociative memory neural network is proposed,with the following characteristics:(1)it can store any given training pattern set no matter how much and what correlation among them may be;(2)the smallest domain of attraction of training patterns is maximized;(3)each domain of attraction of training patterns is guaranteed to be as large as possible;(4)the designed associative memory network is globally stable.A large number of computer experimental results confirm that our algorithm possesses more powerful storage ability and more fault-tolerance capability than existing ones.
Key wordsAssociative memory,training pattern set,total storage ability,maximized domain of attraction,global stability.
1
ɷΪԷԷģͣϢϢĹܣܹӲϢʵϢӦĴ洢еļϢ12.Ҫɾѧϰ㷨.
Ѿ˲ٹѧϰ㷨ҪHebbianѧϰ㷨3ݡͶӰѧϰ㷨4ݡGardnerѧϰ㷨5ݡСصѧϰ㷨6ݡHo-Kashyapѧϰ㷨7ݡԪѵģʽȨѧϰ㷨8ݺŻѧϰ㷨9ݵ.ֻŻѧϰ㷨ϸĴ洢ݴ.
ף9ݵ˼룬ļС׼ҪƳĶԳȨӦʹСļģʽﵽ.һչۺһЧѧϰ㷨ص㣺(1)ȫ洢ѵģʽ(2)Сѵģʽﵽ(3)(2)Ļϣÿѵģʽоܴ(4)ȨԽԪΪ1ĶԳƳȫȶģ11.
2ԷģѵʽĻԼ
ģNԪɵķԶϵͳ.״̬v=(v1,,vN)tvi(i=1,2,,N)ʾiԪ״̬ȡֵռ{-1,1}ȨW=(Wij)NN,WijʾӵjԪiԪȨ.ģͿɱΪ1
