资源描述:
《- Analysis of the dynamical behavior of a feedback》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、ARTICLEINPRESSNeurocomputing71(2008)486–494www.elsevier.com/locate/neucomAnalysisofthedynamicalbehaviorofafeedbackauto-associativememoryMahmoodAmiri,SohrabSaeb,MohammadJavadYazdanpanah,S.AliSeyyedsalehiFacultyofBiomedicalEngineering,AmirkabirUniversityofTechnology(TehranPolytechnic),424Ha
2、fezAve,Tehran15875-4413,IranAvailableonline9October2007AbstractThedynamicalbehaviorandthestabilitypropertiesoffixedpointsinafeedbackauto-associativememoryareinvestigated.Theproposedstructureencompassesamulti-layerperceptron(MLP)andafeedbackconnectionthatlinksinputandoutputlayersthroughdelay
3、elements.TheMLPisinitiallytrainedsothatitmapsthetrainingpatternsintothemselvesasanauto-associativememory.Thefeedbackconnectionisthenestablishedinordertomakethefeedbackauto-associativememory.Wederivesomeexplicitequationsbasedonthetheoryofdynamicalsystems,whichrelatethestabilitypropertiesoffi
4、xedpointstothenetworkparametervalues.Wethenperformsomecasestudiesforthepurposeofperformancecomparisonsbetweentheproposedmodelandaself-feedbackneuralnetwork(SFNN)asanassociativememory.Severalsimulationsareprovidedtoverifythatnotonlyourmodelneedsmuchfewerneuronstostorenumerousstablefixedpoint
5、s,butalsoitisabletolearnasymmetricarrangementoffixedpoints,whereastheSFNNmodelislimitedtoorthogonalarrangements.r2007ElsevierB.V.Allrightsreserved.Keywords:Feedbackauto-associativememory;Stabilityanalysis;Fixedpoints1.Introductionequilibriumpointofthenetwork.Furthermore,theequilibriacorresp
6、ondingtothememoryvectorshavetoUtilizedinawidespectrumofapplications,neuralbeasymptoticallystable,i.e.,theyshouldbeattractivenetworkshavebeenwidelystudiedinrecentyears[8,15,17].equilibriumpointsorattractors[3].IntheconventionalstructureofanartificialneuralAsNNsstoreasetofdesiredpatternsassta
7、blenetwork,aneuronreceivesitsinputeitherfromotherequilibriumpointssuchthatthestoredinformationcanneuronsorfromexternalinputs(inputvector).Aweightedberetrievedifsufficientdataisprovidedinaninputsumoftheseinputsconstitutestheargumentofafixedpattern.Inotherwords,As