1、Y828552∞————一№⑧研究生學(xué)位論文幾類神經(jīng)網(wǎng)絡(luò)的全局指數(shù)穩(wěn)定性——泛函微分方程在神經(jīng)網(wǎng)絡(luò)中自蛙用㈨g割顯里揖目螄hg———二￡——J整—』址——————一q镕監(jiān)Ⅻ亟土々業(yè)名％基園塾雯蝴月M墊堡壘且墨目靴授干日1—一——中國海洋大學(xué)Theglobalexponentialstabilityofseveralkindsofneuralnetworkstheapplicationsoffunctionaldifferentialeq

2、uationinneuralnetworksAbstractDuetothebroadspectrumofapplicationsinoptimization，signalprocessing，imageprocessing，solvingnonlinearalgebraicequation，patternrecognition，associativememoriesandSOon，theartfficialneuralnetworks

3、(ANN)havebeenfullydevelopedBecausethefunctionsininformationprocessingisdeterminedbythedynamicqualityoftheartificialneuralnetworks，itisnecessarytoinvestigatedthedynamicqualitiessuchasstability，oscillation，bifurcationandch

4、aosandSOonEspecially，intheapplicationsofoptimization，neuralcontrol，signalprocessingandpatternrecognition，theglobalasymptoticstabifityhasbecomethecoreproblemintheresearchinordertoavoidthespuriousresponseorlocalminimaInthe

5、realizationofelectronicneuralnetworks，timedelaysarelikelytobepresentduetothefiniteswitchingspeedofamplifiers，whichwillaffectthestabilityandleadtotheoscillationandthereforeinfluencethefunctionoftheartificialneuralnetworks

6、So，thethedynamicbehaviorofthetwokindsofANNmodelswithdelaysandwithoutdelayshavebeeninvestigatedinthisthesisThefundamentalproblemsoftheANNdynamicmodelsarethefollowing：1theexistenceofequihbrium；2theattractivity；3thestabilit

7、yInthisthesisthefeasiblecriteriaofseveralANNstocriticizeexistenceoftheuniqueglobalexponentialstableequilibriumInthisthesis，therefor，anewmethodofanalyzingthegloballyexponentialstabilityofHopfieldNeuralNetworksispresentedi

8、nchapter1，andthecriteriumofthegloballyexponentialstabilityofBidirectionalAssociativeMemoryNeu—ralNetworkswithVaryingTimeDelaysisalsoacquiredinchapter2Besides，anewlemmaabouttheHaianayDifferentialInequalityisobtainedandane