资源描述：
《- Improved_chaotic_associative_memory_for_successive_learning》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、14ImprovedChaoticAssociativeMemoryforSuccessiveLearningTakahiroIKEYAandYukoOSANATokyoUniversityofTechnologyJapan1.IntroductionRecently,neuralnetworksaredrawingmuchattentionasamethodtorealizeflexibleinformationprocessing.Neuralnetworksconsiderneurongroupsofthebraininthecreature,andimitatet
2、heseneuronstechnologically.Neuralnetworkshavesomefeatures,especiallyoneoftheimportantfeaturesisthatthenetworkscanlearntoacquiretheabilityofinformationprocessing.Inthefiledofneuralnetwork,manymodelshavebeenproposedsuchastheBackPropagationalgorithm(Rumelhartetal.,1986),theSelf-OrganizingMap
3、(Kohonen,1994),theHopfieldnetwork(Hopfield,1982)andtheBidirectionalAssociativeMemory(Kosko,1988).Inthesemodels,thelearningprocessandtherecallprocessaredivided,andthereforetheyneedallinformationtolearninadvance.However,intherealworld,itisverydifficulttogetallinformationtolearninadvance.Sow
4、eneedthemodelwhoselearningandrecallprocessesarenotdivided.Assuchmodel,GrossbergandCarpenterproposedtheAdaptiveResonanceTheory(ART)(Carpenter&Grossberg,1995).However,theARTisbasedonthelocalrepresentation,andthereforeitisnotrobustfordamage.Whileinthefieldofassociativememories,somemodelshave
5、beenproposed(Watanabeetal.,1995;Osana&Hagiwara,1999;Kawasakietal.,2000;Ideguchietal.,2005).Sincethesemodelsarebasedonthedistributedrepresentation,theyhavetherobustnessfordamagedneurons.However,theirstoragecapacityisverysmallbecausetheirlearningprocessesarebasedontheHebbianlearning.Incontr
6、ast,theHeteroChaoticAssociativeMemoryforSuccessiveLearningwithgiveupfunction(HCAMSL)(Arai&Osana,2006)andtheHeteroChaoticAssociativeMemoryforSuccessiveLearningwithMulti-Winnerscompetition(HCAMSL-MW)(Andoetal.,2006)havebeenproposedinordertoimprovethestoragecapacity.Inthisresearch,weproposea
7、nImprovedChaoticAssociativeMemoryforSuccessiveLearning(ICAMSL).TheproposedmodelisbasedontheHeteroChaoticAssociativeMemoryforSuccessiveLearningwithgiveupfunction(HCAMSL)(Arai&Osana,2006)andtheHeteroChaoticAssociativeMemoryforSuccessiveLearningwithMulti-Winnerscompeti
