人工神经网络及应用智慧树知到课后章节答案2023年下长安大学.docx

人工神经网络及应用智慧树知到课后章节答案2023年下长安大学长安大学

第一章测试

Synapseistheplacewhereneuronsconnectinfunction.Itiscomposedofpresynapticmembrane,synapticspaceandpostsynapticmembrane.（）

A:对B:错

答案:对

Biologicalneuronscanbedividedintosensoryneurons,motorneuronsand（）accordingtotheirfunctions.

A:multipolarneuronsB:interneurons

C:PseudounipolarneuralnetworksD:bipolarneurons

答案:interneurons

Neuronsandglialcellsarethetwomajorpartsofthenervoussystem.（）

A:错B:对

答案:对

Neuronsarehighlypolarizedcells,whicharemainlycomposedoftwoparts:thecellbodyandthesynapse.（）

A:错B:对

答案:对

Thehumanbrainisanimportantpartofthenervoussystem,whichcontainsmorethan86billionneurons.Itisthecentralinformationprocessingorganizationofhumanbeings.（）

A:对B:错

答案:对

第二章测试

In1989,Mead,thefatherofVLSI,publishedhismonograph"()",inwhichageneticneuralnetworkmodelbasedonevolutionarysystemtheorywasproposed.

A:LearningMachines

B:JournalNeuralNetworks

C:AnalogVLSIandNeuralSystems

D:Perceptrons:AnIntroductiontoComputationalGeometry

答案:AnalogVLSIandNeuralSystems

In1989,YannLecunproposedconvolutionalneuralnetworkandappliedittoimageprocessing,whichshouldbetheearliestapplicationfieldofdeeplearningalgorithm.（）

A:对B:错

答案:对

In1954,Eccles,aneurophysiologistattheUniversityofMelbourne,summarizedtheprincipleofDale,aBritishphysiologist,that"eachneuronsecretesonlyonekindoftransmitter".（）

A:错B:对

答案:对

In1972,ProfessorKohonenofFinlandproposedaself-organizingfeaturemap(SOFM)neuralnetworkmodel.（）

A:对B:错

答案:对

Predictionandevaluationisanactivityofscientificcalculationandevaluationofsomecharacteristicsanddevelopmentstatusofthingsoreventsinthefutureaccordingtotheknowninformationofobjectiveobjects.（）

A:对B:错

答案:对

第三章测试

Thefunctionoftransferfunctioninneuronsistogetanewmappingoutputofsummeraccordingtothespecifiedfunctionrelationship,andthencompletesthetrainingofartificialneuralnetwork.（）

A:对B:错

答案:对

Thedeterminantchangessignwhentworows(ortwocolumns)areexchanged.Thevalueofdeterminantiszerowhentworows(ortwocolumns)aresame.（）

A:对B:错

答案:对

Therearetwokindsofphenomenaintheobjectiveworld.Thefirstisthephenomenonthatwillhappenundercertainconditions,whichiscalledinevitablephenomenon.Thesecondkindisthephenomenonthatmayormaynothappenundercertainconditions,whichiscalledrandomphenomenon.（）

A:错B:对

答案:对

LogarithmicS-typetransferfunction,namelySigmoidfunction,isalsocalledS-shapedgrowthcurveinbiology.（）

A:错B:对

答案:对

Rectifiedlinearunit(ReLU),similartotheslopefunctioninmathematics,isthemostcommonlyusedtransferfunctionofartificialneuralnetwork.（）

A:错B:对

答案:对

第四章测试

Theperceptronlearningalgorithmisdrivenbymisclassification,sothestochasticgradientdescentmethodisusedtooptimizethelossfunction.（）

A:misclassificationB:maximumC:minimum

D:correct

答案:misclassification

Perceptronisasingle-layerneuralnetwork,orneuron,whichisthesmallestunitofneuralnetwork.（）

A:错B:对

答案:对

Whentheperceptronislearning,eachsamplewillbeinputintotheneuronasastimulus.Theinputsignalisthefeatureofeachsample,andtheexpectedoutputisthecategoryofthesample.Whentheoutputisdifferentfromthecategory,wecanadjustthesynapticweightandbiasvalueuntiltheoutputofeachsampleisthesameasthecategory.（）

A:对B:错

答案:对

Ifthesymmetrichardlimitfunctionisselectedforthetransferfunction,theoutputcanbeexpressedas.Iftheinnerproductoftherowvectorandtheinputvectorintheweightmatrixisgreaterthanorequalto-b,theoutputis1,otherwisetheoutputis-1.（）

A:错B:对

答案:对

Thebasicideaofperceptronlearningalgorithmistoinputsamplesintothenetworkstepbystep,andadjusttheweightmatrixofthenetworkaccordingtothedifferencebetweentheoutputresultandtheidealoutput,thatistosolvetheoptimizationproblemoflossfunctionL(w,b).（）

A:错B:对

答案:对

第五章测试

TheoutputofBPneuralnetworkis（）ofneuralnetwork.

A:theoutputofthelastlayerB:theinputofthelastlayer

C:theoutputofthesecondlayerD:theinputofthesecondlayer

答案:theoutputofthelastlayer

BPneuralnetworkhasbecomeoneofthemostrepresentativealgorithmsinthefieldofartificialintelligence.Ithasbeenwidelyusedinsignalprocessing,patternrecognition,machinecontrol(expertsystem,datacompression)andotherfields.（）

A:对B:错

答案:对

In1974,PaulWerbosofthenaturalsciencefoundationoftheUnitedStatesfirstproposedtheuseoferrorbackpropagationalgorithmtotrainartificialneuralnetworksinhisdoctoraldissertationofHarvardUniversity,anddeeplyanalyzedthepossibilityofapplyingittoneuralnetworks,effectivelysolvingtheXORloopproblemthatsinglesensorcannothandle.（）

A:对B:错

答案:对

InthestandardBPneuralnetworkalgorithmandmomentumBPalgorithm,thelearningrateisaconstantthatremainsconstantthroughoutthetrainingprocess,andtheperformanceofthelearningalgorithmisverysensitivetotheselectionofthelearningrate.（）

A:对B:错

答案:对

L-Malgorithmismainlyproposedforsuperlargescaleneuralnetwork,anditisveryeffectiveinpracticalapplication.（）

A:对B:错

答案:错

第六章测试

RBFneuralnetworkisanovelandeffectivefeedforwardneuralnetwork,whichhasthebestlocalapproximationandglobaloptimalperformance.（）

A:对B:错

答案:对

Atpresent,RBFneuralnetworkhasbeensuccessfullyappliedinnonlinearfunctionapproximation,timeseriesanalysis,dataclassification,patternrecognition,informationprocessing,imageprocessing,systemmodeling,controlandfaultdiagnosis.（）

A:对B:错

答案:对

ThebasicideaofRBFneuralnetworkistouseradialbasisfunctionasthe"basis"ofhiddenlayerhiddenunittoformhiddenlayerspace,andhiddenlayertransformsinputvector.Theinputdatatransformationoflowdimensionalspaceismappedintohigh-dimensionalspace,sothattheproblemoflinearseparabilityinlow-dimensionalspacecanberealizedinhigh-dimensionalspace.（）

A:对B:错

答案:对

ForthelearningalgorithmofRBFneuralnetwork,thekeyproblemistodeterminethecenterparametersoftheoutputlayernodereasonably.（）

A:对B:错

答案:错

ThemethodofselectingthecenterofRBFneuralnetworkbyself-organizinglearningistoselectthecenterofRBFneuralnetworkbyk-meansclusteringmethod,whichbelongstosupervisedlearningmethod.（）

A:错B:对

答案:错

第七章测试

Intermsofalgorithm,ADALINEneuralnetworkadoptsW-Hlearningrule,alsoknownastheleastmeansquare(LMS)algorithm.Itisdevelopedfromtheperceptronalgorithm,anditsconvergencespeedandaccuracyhavebeengreatlyimproved.（）

A:错B:对

答案:对

ADALINEneuralnetworkhassimplestructureandmulti-layerstructure.Itisflexibleinpracticalapplicationandwidelyusedinsignalprocessing,systemidentification,patternrecognitionandintelligentcontrol.（）

A:对B:错

答案:对

WhentherearemultipleADALINEinthenetwork,theadaptivelinearneuralnetworkisalsocalledMadalinewhichmeansmanyAdalineneuralnetworks.（）

A:对B:错

答案:对

Thealgorithmusedinsingle-layerADALINEnetworkisLMSalgorithm,whichissimilartothealgorithmofperceptron,andalsobelongstosupervisedlearningalgorithm.（）

A:对B:错

答案:对

Inpracticalapplication,theinverseofthecorrelationmatrixandthecorrelationcoefficientarenoteasytoobtain,sotheapproximatesteepestdescentmethodisneededinthealgorithmdesign.Thecoreideaisthattheactualmeansquareerrorofthenetworkisreplacedbythemeansquareerrorofthek-thiteration.

（）

A:错B:对

答案:对

第八章测试

Hopfieldneuralnetworkisakindofneuralnetworkwhichcombinesstoragesystemandbinarysystem.Itnotonlyprovidesamodeltosimulatehumanmemory,butalsoguaranteestheconvergenceto（）.

A:localminimumB:localmaximum

C:minimum

D:maximum

答案:localminimum

Atpresent,researchershavesuccessfullyappliedHopfieldneuralnetworktosolvethetravelingsalesmanproblem(TSP),whichisthemostrepresentativeofoptimizationcombinatorialproblems.（）

A:错B:对

答案:对

In1982,AmericanscientistJohnJosephHopfieldputforwardakindoffeedbackneuralnetwork"Hopfieldneuralnetwork"inhispaperNeuralNetworksandPhysicalSystemswithEmergentCollectiveComputationalAbilities.（）

A:对B:错

答案:对

Undertheexcitationofinputx,DHNNentersadynamicchangeprocess,untilthestateofeachneuronisnolongerchanged,itreachesastablestate.Thisprocessisequivalenttotheprocessofnetworklearningandmemory,andthefinaloutputofthenetworkisthevalueofeachneuroninthestablestate.（）

A:错B:对

答案:对

Theorderinwhichneuronsadjusttheirstatesisnotunique.Itcanbeconsideredthatacertainordercanbespecifiedorselectedrandomly.Theprocessofneuronstateadjustmentincludesthreesituations:from0to1,and1to0andunchanged.（）

A:错B:对

答案:对

第九章测试

ComparedwithGPU,CPUhashigherprocessingspeed,andhassignificantadvantagesinprocessingrepetitivetasks.（）

A:对B:错

答案:错

Atpresent,DCNNhasbecomeoneofthecorealgorithmsinthefieldofimagerecognition,butitisunstablewhenthereisasmallamountoflearningdata.（）

A:对B:错

答案:错

Inthefieldoftargetdetectionandclassification,thetaskofthelastlayerofneuralnetworkistoclassify.（）

A:对B:错

答案:对

InAlexNet,thereare650000neuronswithmorethan600000parametersdistributedinfiveconvolutionlayersandthreefullyconnectedlayersandSoftmaxlayerswith1000categories.（）

A:对B:错

答案:错

VGGNetiscomposedoftwoparts:theconvolutionlayerandthefullconnectionlayer,whichcanberegardedasthedeepenedversionofAlexNet.（）

A:错B:对

答案:对

第十章测试

TheessenceoftheoptimizationprocessofDandGistofindthe（）.

A:maximumB:minimaxC:localmaxima

D:minimum

答案:minimax

Intheartificialneuralnetwork,thequalityofmodelingwilldirectlyaffecttheperformanceofthegenerativemodel,butasmallamountofpriorknowledgeisneededfortheactualcasemodeling.（）

A:对B:错

答案:错

AGANmainlyincludesageneratorGandadiscriminatorD.（）

A:对B:错

答案:对

Becausethegenerativeadversarialnetworkdoesnotneedtodistinguishthelowerboundandapproximateinference,itavoidsthepartitionfunctioncalculationproblemcausedbythetraditionalrepeatedapplicationofMarkovchainlearningmechanism,andimprovesthenetworkefficiency.（）

A:对B:错

答案:对

Fromtheperspectiveofartificialintelligence,GANusesneuralnetworktoguideneuralnetwork,andtheideaisverystrange.（）

A:对B:错

答案:对

第十一章测试

ThecharacteristicofElmanneuralnetworkisthattheoutputofthehiddenlayerisdelayedandstoredbythefeedbacklayer,andthefeedbackisconnectedtotheinputofthehiddenlayer,whichhasthefunctionofinformationstorage.（）

A:对B:错

答案:对

InElmannetwork,thetransferfunctionoffeedbacklayerisnonlinearfunction,andthetransferfunctionofoutputlayerislinearfunction.（）

A:对B:错

答案:对

Thefeedbacklayerisusedtomemorizetheoutputvalueoftheprevioustimeofthehiddenlayerunitandreturnittotheinput.Therefore,Elmanneuralnetworkhasdynamicmemoryfunction.（）

A:对B:错

答案:对

TheneuronsinthehiddenlayerofElmannetworkadoptthetangentS-typetransferfunction,whiletheoutputlayeradoptsthelineartransferfunction.Ifthereareenoughneuronsinthefeedbacklayer,thecombinationofthesetransferfunctionscanmakeElmanneuralnetworkapproachanyfunctionwitharbitraryprecisioninfinitetime.（）

A:对B:错

答案:对

Elmanneuralnetworkisakindofdynamicrecurrentnetwork,whichcanbedividedintofullfeedbackandpartialfeedback.Inthepartialrecurrentnetwork,thefeedforwardconnectionweightcanbemodified,andthefeedbackconnectioniscomposedofagroupoffeedbackunits,andtheconnectionweightcannotbemodified.（）

A:错B:对

答案:对

第十二章测试

ThelossfunctionofAdaBoostalgorithmis（）.

A:exponentialfunction

B:nonlinearfunctionC:linearfunction

D:logarithmicfunction

答案:exponentialfunction

Boostingalgorithmisthegeneralnameofaclassofalgorithms.Theircommongroundistoconstructastrongclassifierbyusingagroupofweakclassifiers.Weakclassifiermainlyreferstotheclassifierwhosepredictionaccuracyisnothighandfarbelowtheidealclassificationeffect.Strongclassifiermainlyreferstotheclassifierwithhighpredictionaccuracy.（）

A:错B:对

答案:对

Amongthemanyimprovedboostingalgorithms,themostsuccessfuloneistheAdaBoost(adaptiveboosting)algorithmproposedbyYoavFreundofUniversityofCaliforniaSanDiegoandRobertSchapireofPrincetonUniversityin1996.（）

A:错B:对

答案:对

ThemostbasicpropertyofAdaBoostisthatitreducesthetrainingerrorcontinuouslyinthelearningprocess,thatis,theclassificationerrorrateonthetrainingdatasetuntileachweakclassifieriscombinedintothefinalidealclassifier.（）

A:错B:对

答案:对

ThemainpurposeofaddingregularizationtermintotheformulaofcalculatingstrongclassifieristopreventtheoverfittingofAdaBoostalgorithm,whichisusuallycalledstepsizeinalgorithm.（）

A:错B:对

答案:对

第十三章测试

ThecorelayerofSOFMneuralnetworkis（）.

A:inputlayerB:hiddenlayer

C:outputlayerD:competitionlayer

答案:competitionlayer

Inordertodividetheinputpatternsintoseveralclasses,thedistancebetweeninputpatternvectorsshouldbemeasuredaccordingtothesimilarity.（）areusuallyused.

A:EuclideandistancemethodB:Cosinemethod

C:SinemethodD:Euclideandistancemethodandcosinemethod

答案:Euclideandistancemethodandcosinemethod

SOFMneuralnetworksaredifferentfromotherartificialneuralnetworksinthattheyadoptcompetitivelearningratherthanbackwardpropagationerrorcorrectionlearningmethodsimilartogradientdescent,andinasense,theyuseneighborhoodfunctionstopreservetopologicalpropertiesofinputspace.（）

A:对B:错

答案:对

ForSOFMneuralnetwork,thecompetitivetransferfunction(CTF)responseis0forthewinningneurons,and1forotherneurons.（）

A:错B:对

答案:错

Whentheinputpatterntothenetworkdoesnotbelongtoanypatterninthenetworktrainingsamples,SOFMneuralnetworkcanonlyclassifyitintotheclosestmode.（）

A:对B:错

答案:对

第十四章测试

Theneuralnetworktoolboxcontains（）modulelibraries.

A:threeB:six

C:fiveD:four

答案:five

The"netprod"inthenetworkinputmodulecanbeusedfor（）.

A:dotmultiplicationB:dotdivision

C:additionorsubtraction

D:dotmultiplicationordotdivision

答案:dotmultiplicationordotdivision

The"dotrod"intheweightsettingmoduleisanormaldotproductweightfunction.（）

A:错B:对

答案:错

Themathematicalmodelofsingleneuronisy=f(wx+b).（）

A:错B:对

答案:对

Theneuronmodelcanbedividedintothreeparts:inputmodule,transferfunctionandoutputmodule.（）

A:对B:错

答案:对

第十五章测试

Inlarge-scalesystemsoftware