欢迎来到天天文库
浏览记录
ID:42102918
大小:159.27 KB
页数:5页
时间:2019-09-08
《人工神经网络试题及答案》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、QuestionOne:TheweightupdatingrulesoftheperceptronandKohonenneuralnetworkarew(n+1)=w(n)+r
2、d(n)x(n);x(n)=inputvector=[+1Jx1(n),x2(n),...,xm(n)Fw(n)=weightvector=[b(n),w^n),w2(n),wm(n)卩b(n)二biasy(n)二actualresponsed(n)=desiredresponser]=learningrateparameter+1)=Wy(w)+7(w)方心)
3、a(6)(X(n)-W丿⑷)QuestionTwo:Thelimitationoftheperceptronisthatitcanonlymodellinearlyseparableclasses.ThedecisionboundaryofRBFislinearwhereasthedecisionboundaryofFFNNisnon-linear•QuestionThree:TheactivationfunctionoftheneuronofthePerceptron,BPnetworkandRBFnetworkarerespecti
4、vely;;.0(v)=+lifv>0-1ifv<0witha>0—av・l+€丿Gaussian尸20("=exp(—丁万)NN3forsomeQ>0QuestionFour:Pleasepresenttheidea,objectivefunctionoftheBPneuralnetworks(FFNN)andthelearningruleoftheneuronattheoutputlayerofFFNN.Youareencouragedtowritedowntheprocesstoproducethelearningrule.wit
5、hq>0Weightupdateofoutputneuron3•・=jbecauseIfjisanoutputneuronthenusingthechainruleweobtain:dEdEde■dy■7=--_e八一I)©(vJon.ejyjej=dj-yjandForjoutputneuron巧予0(Vj)SubstitutinginAwji=wegetAw〃=〃(dj・yj)©,(v丿)儿QuestionFive:PleasedescribethesimilarityanddifferencebetweenHopfieldNNan
6、dBoltzmannmachine.相同:Bothofthemaresingle-layerinter-connectionNNs.Theybothhavesymmetricweightmatrixwhosediagonalelementsarezeroes.不同:ThenumberoftheneuronsofHopfieldNNisthesameasthenumberofthedimension(K)ofthevectordata.Ontheotherhand,BoltzmannmachinewillhaveK+Lneurons.Th
7、erearcLhiddenneuronsBoltzmannmachinehasKneuronsthatservesasbothinputneuronsandoutputneurons(Auto-associationBoltzmannmachine)・QuestionSix:Pleaseexplainthetermsintheaboveequationindetail.PleasedescribetheweightupdatingequationsofeachnodeinthefollowingFFNNusingtheBPlearnin
8、galgorithm.(PPT原题y=(p(net)=(p(wo+wiXi+w2X2))aw2)15(^(w0+WjXj+w2x2)-dy20W]=丄29(%+忖+柿2)-严(%+讣+讣)-”)>2*zz、八—、0(叫+WE+%兀2一”)二0(/7/)—)」=((p(net)_d)0■Aw0=〃(d-(p{nety)0{net)Aw】=r/(d-(p{net))(pet)xtAw2=〃(d-(p{net))(pet)x2AwnW0=w0+Uwz+ZW2=w2+.LQuestionSeven:Pleasetryyourbestt
9、opresentthecharacteristicsofRBFNN.(1)RBFnetworkshaveonesinglehiddenlayer・(2)InRBFtheneuronmodelofthehid
此文档下载收益归作者所有