资源描述：
《preventing computational chaos in asynchronous neural networksnew》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。
1、Preprint,InternationalConferenceonArtificialIntelligenceApplicationsMalaga,Spain,September2002AIA’02PreventingComputationalChaosinAsynchronousNeuralNetworksJacobBarhenVladimirProtopopescuCenterforEngineeringScienceAdvancedResearchComputingandComputationalSciencesDirectora
2、teOakRidgeNationalLaboratoryOakRidge,TN37831-6355barhenj@ornl.govAbstract.Oneoftheprimaryadvantagesofartificialfromallrequirednodes.Clearly,thistypeofasyn-neuralnetworksistheirinherentabilitytoperformchronicitylimitstheabilityofanetworktoperformmassivelyparallel,nonlinear
3、signalprocessing.massivelyparallel,distributedinformationpro-However,theasynchronousdynamicsunderlyingthecessing.Hereafter,wewillrefertothisregimeasevolutionofsuchnetworksmayoftenleadtothesequentiallyasynchronous.Todate,bothparadigmsemergenceofcomputationalchaos,whichimpe
4、desstillprovidethealgorithmicfoundationofavailabletheefficientretrievalofinformationusuallystoredincomputationalmodels[3-6].thesystem’sattractors.Inthispaper,wediscusstheThetrue(concurrent)computationalasynchronicity,implicationsofchaosinconcurrentasynchronoushowever,impl
5、iesanuncoordinated,system−widecomputation,andprovideamethodologythatactivity.Inthatcontext,thereisastrongmotivationpreventsitsemergence.Ourresultsareillustratedontodevelopalgorithmsthatcanfullyexploitsuchaawidelyusedneuralnetworkmodel.behavior.Itshouldbenoted,however,that
6、asynchronousrelaxationalgorithmshavelongbeenknowntogiverisetocomputationalchaos[10].1.IntroductionInthesequel,wefirstdiscusssomeimplicationsofArtificialneuralnetworksaremassivelyparallel,asynchronouscomputing.Thenweprovideametho-adaptivedynamicalsystems[1].Theirmodelsared
7、ologythatpreventstheemergenceofcomputationalinspiredbythegeneralfeaturesofbiologicalchaostoenableefficientretrievalofinformationnetworks.Insuchnetworks,asynchronousbehaviorstoredinattractorsofthenetwork.Finally,weisprevalent.Itarisesfromdelaysinnervesignalillustrateourres
8、ultsintermsofthewellestablishedpropagation,refractoryperiods,andadaptiveGrossberg−Hopfieldmodel[
