Influence of Parameter Deviations in An

《Influence of Parameter Deviations in An》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、InfluenceofParameterDeviationsinAnAssociativeChaoticNeuralNetworkMasaharuADACHIDepartmentofElectronicEngineering,SchoolofEngineeringTokyoDenkiUniversityTokyo101-8457,JapanE-mail:adachi@d.dendai.acJpAbstract-Influenceofparameterdeviationsinanassociativenetworkstheinfluenceoftheparameterdev

2、iationscanbechaoticneuralnetworkisanalyzedinthispaper.Thenetworkneglectedsincetheconstituentneuronsarejustnonlineariscomposedofchaoticmodelneuronswhoseparametervaluesmappingdevices(e.g.,thresholddevices).Ontheotherhandhavesmalldeviationsamongtheneuronsinthenetwork.Theassociativechaoticneu

3、ralnetworkwith16chaoticneuronsistheinfluenceinthechaoticneuralnetworkscannotbeseenanalyzedinthispaper.Threeorthogonalpatternvectorsaresimpleasintheconventionalonebecausetheconstituentstoredbysettingsynapticweightsofthenetworkwithconven-neuronsmayexhibitchaoticdynamics.Thereforeitisworthti

4、onalauto-associativematrix.Comparisonsofthenetworkwithtoinvestigateontheinfluenceoftheparameterdeviationsparameterdeviationsandthenetworkwithcommonparametersinthechaoticneuralnetworksbothinbrainscienceandaremadeintheretrievalcharacteristicsandinthedynamicalproperty.Inordertomakesuchcompar

5、ison,thenominalvaluesinengineering.Inthispaper,anassociativechaoticneuraloftheparametersofthenetworkaresetsothatthenetworksnetworkisanalyzedasanexample.showsperiodicretrievalofoneofthestoredpatternswhentheparametervaluesarecommontoeveryneuroninthenetwork.11.ASSOCIATIVECHAOTICNEURALNETWORK

6、ThenetworkwiththeparameterdeviationsamongtheneuronsTheoperationoftheassociativechaoticneuralnetworkshowsaperiodicassociativedynamicsevenifthenominalvaluesoftheparametersaresettoshowperiodicbehavior.Themodelisrepresentedbythefollowingequations:aperiodicassociativedynamicsofthenetworkwithth

7、eparameterdeviationscanhecharacterizedbytheindexfortheorbitalXi(t+1)=f{,(t+1)+6(t+I)},(1)instability.16I.INTRODUCTION?<(t+1)=kp7Ji(t)+CWijZj(t),(2)Achaoticneuralnetwork,whichiscomposedofchaoticj=lneuronmodels,wasproposed[I@].Thechaoticneuronmodel<<(t+1)=kt')