人工神经网络试题及答案

QuestionOne:TheweightupdatingrulesoftheperceptronandKohonenneuralnetworkare_____.QuestionTwo:Thelimitationoftheperceptronisthatitcanonlymodellinearlyseparableclasses.ThedecisionboundaryofRBFis__________linear______________________whereasthedecisionboundaryofFFNNis__________________non-linear___________________________.QuestionThree:TheactivationfunctionoftheneuronofthePerceptron,BPnetworkandRBFnetworkarerespectively________________;________________;______________.QuestionFour:Pleasepresenttheidea,objectivefunctionoftheBPneuralnetworks(FFNN)andthelearningruleoftheneuronattheoutputlayerofFFNN.Youareencouragedtowritedowntheprocesstoproducethelearningrule.QuestionFive:PleasedescribethesimilarityanddifferencebetweenHopfieldNNandBoltzmannmachine.相同：Bothofthemaresingle-layerinter-connectionNNs.Theybothhavesymmetricweightmatrixwhosediagonalelementsarezeroes.不同：ThenumberoftheneuronsofHopfieldNNisthesameasthenumberofthedimension(K)ofthevectordata.Ontheotherhand,BoltzmannmachinewillhaveK+Lneurons.ThereareLhiddenneuronsBoltzmannmachinehasKneuronsthatservesasbothinputneuronsandoutputneurons(Auto-associationBoltzmannmachine).QuestionSix:Pleaseexplainthetermsintheaboveequationindetail.PleasedescribetheweightupdatingequationsofeachnodeinthefollowingFFNNusingtheBPlearningalgorithm.(PPT原题y=φ(net)=φ(w0+w1x1+w2x2))W0=w0+W1=w1+W2=w2+QuestionSeven:PleasetryyourbesttopresentthecharacteristicsofRBFNN.(1)RBFnetworkshaveonesinglehiddenlayer.(2)InRBFtheneuronmodelofthehiddenneuronsisdifferentfromtheoneoftheoutputnodes.(3)ThehiddenlayerofRBFisnon-linear,theoutputlayerofRBFislinear.(4)TheargumentofactivationfunctionofeachhiddenneuroninaRBFNNcomputestheEuclideandistancebetweeninputvectorandthecenterofthatunit.(5)RBFNNusesGaussianfunctionstoconstructlocalapproximationstonon-linearI/Omapping.QuestionEight:Generally,theweightvectorsofallneuronsofSOMisadjustedintermsofthefollowingrule:wj(n+1)=wj(n)+η(n)hi(x)(di(x)j)(x(n)-wj(n)).Pleaseexplaineachtermintheaboveformula.:weightvalueofthej-thneuronatiterationn:neighborhoodfunctiondji:lateraldistanceofneuronsiandj:thelearningrate:thewinningneuronmostadjacenttoXX:oneinputexample