人工神經(jīng)網(wǎng)絡(luò)及應(yīng)用智慧樹知到期末考試答案+章節(jié)答案2024年長安大學(xué)ThecommonlyusedtransferoractivationfunctionsoffullyconnectedlayerincludeSigmoidandSoftmax.（）

答案:對In1976,Grossberg,acognitiveneuroscientistatBostonUniversity,proposedamoreperfectperceptronmodel,adaptiveresonancetheory(ART),thatis,supervisedlearning.（）

答案:對Throughrepeatedlearning,SOFMneuralnetworkcanmakethespatialdistributiondensityofconnectionweightsconsistentwiththeprobabilitydistributionofinputpatterns,soitcancompletetheclassificationofinputpatternsthroughthespatialdistributionofconnectionweights.（）

答案:對Agradientisavectorthatindicatesthatthedirectionalderivativeofafunctionatacertainpointreachesitsmaximumvalueinthatdirection.（）

答案:對Intheearly1990s,Edelman,theNobellaureate,putforwardthreeformsofDarwinismmodelandestablishedaneuralnetworksystemtheoryinvolvinggrouprestriction,groupselectionandgroupcompetition.（）

答案:對Amari,aJapaneseresearcher,describedtrustallocationinbiologicalneuralnetworkswithmathematicallanguage.（）

答案:對AftertheElmanneuralnetworkisestablishedinMATLAB,theweightsandbiasvaluesofeachlayerareinitializedaccordingtoNguyenWidrowalgorithm,andtheinitializationfunctionisinitnw.（）

答案:對In1994,Hebb,aneuropsychologist,proposedtheearliestlearningrulesforneuralnetworks,whichwereusedtoadjusttheweightsbetweenneurons.（）

答案:對Rosenblattisoneofthemainconstructorsofmodernneuralnetwork.Hisresearchhasarousedgreatinterestandattentionofmanyscholarsandorganizations.（）

答案:對Biologicalstudieshaveshownthatneuronsarearrangedinanorderlymannerinthesensorychannelsofthehumanbrain.Neuronsindifferentregionshavedifferentfunctionsanddifferentinputinformationpatterns.（）

答案:對FERNNalgorithmcanextractMOFMrulesorDNFrulesaftertrainingneuralnetworkformanytimes.（）

答案:對Intheprocessofelectricalsynaptictransmission,chemicalsubstancesaretransmittedthroughgapjunctionsbetweentheanteriorandposteriorneurons.Thegapisusuallyverysmall,only2-3cm.()

答案:錯OnJune23,1960,Rosenblatt,apsychologistatCornellUniversityintheUnitedStates,developedaperceptronbasedneuralcomputermark1,whichcanrecognizeEnglishletters.（）

答案:對Actionpotential(AP)isaspikelikepotentialchangeonthebasisofrestingmembranepotentialafterneuronsarestimulated.Itisthebasiclanguageofinformationtransmissioninthenervoussystem.（）

答案:對Overfittingmeansthatthemodelselectedduringlearningcontainstoofewparameters,thatis,thenumberofsamplesisverysmall.（）

答案:錯Facedetectionandfacerecognitionarethesameconcept,whichcanbeusedtodetectthefaceinthescene.（）

答案:錯Thecellmembraneonthesurfaceofaxoniscalledaxon,andthecytoplasminsideiscalledaxon.（）

答案:對Thepoolingoperationcanbedividedintotwoways:averagepoolingandmaximumpooling.（）

答案:對Hopfieldneuralnetworkisasetofnonlineardifferentialequations.TheLyapunovfunctionisintroducedintothenetworkforthefirsttime,thusthestabilityofthenetworkisproved.（）

答案:對Hopfieldneuralnetworkisakindoffeedbackneuralnetwork,whichhasthefunctionoflearningandmemory.Stabilityisoneofthemostimportantcharacteristicsofthisneuralnetworkmodel.（）

答案:對Hopfieldneuralnetworkisakindoffeedbacknetworkwithassociativememoryfunction.Itcanbedividedintodiscretetypeandcontinuoustype.（）

答案:對ADCNNwithcompletefunctionsisusuallycomposedofinputlayer，hiddenlayer,outputlayerorclassificationlayer.（）

答案:對ThearchitectureofRBFneuralnetworkissimilartothemultilayerforwardnetwork,whichisathree-layerforwardneuralnetworkwithsinglehiddenlayer.Theinputlayeriscomposedofsignalsourcenodes,andthehiddenlayerisasingleneuronlayer.（）

答案:對SOFMneuralnetworkformsself-organizingmappingspacebyintroducinggrid,andestablishestopologicalconnectionamongneurons.（）

答案:對RBFneuralnetworkisakindofneuralnetworkstructurewithgoodlocalapproximationperformance.Somescholarshaveprovedthatitcanapproachanycontinuousfunctionswitharbitraryprecision.（）

答案:對Whentheerrorbetweentheactualoutputandtheidealoutputexceedstheexpectedvalue,itisnecessarytoentertheerrorback-propagationprocess.（）

答案:對Whenalltheelementsinthematrixarezero,thematrixiscalledidentitymatrix.（）

答案:錯Theideaoforthogonalleastsquare(OLS)comesfromlinearregressionmodel.（）

答案:對Ionpumpisoneofmembranetransportproteins,whichcandriveionsacrosscellmembraneinreverseconcentrationgradient.（）

答案:對DifferentfromthealgorithmderivationofBPnetwork,recursivenetworkgenerallyadoptsthealgorithmoforderedchainrule.（）

答案:對Deepconvolutionalgenerativeadversarialnetworks(DCGAN)istoaddconvolutionalneuralnetworktothegenerationmodelanddiscriminantmodel,whichgreatlyimprovesthegenerationperformance.（）

答案:對8.Forthelearningeffectofthesametrainingset,differentstepsizemeansdifferentiterationtimesofweakclassifier,andthelargerstepsizemeansmoreiterationtimesofweakclassifier.（）

答案:錯ADALINEneuralnetworkreflectsthelinearmappingrelationshipintheinputandoutputsamplevectorspace.ItusesLMSalgorithm,andthemeansquareerrorofthenetworkpresentsaparabolicsurfaceofquadraticfunction,andthevertexofthesurfaceistheoptimalsolutionoftheneuralnetwork.（）

答案:對Inputlayergenerallyreferstotheneuralnetworklayerusedforinputimage.Thehiddenlayerincludestheconvolutionallayer,thepoolinglayerandthefullyconnectedlayer.（）

答案:對Adaptivelinearneuron(ADALINE)wasproposedbyStanfordProfessorWidrowandhisgraduatestudentHoffin1960.Itwasalmostatthesametimeastheperceptron.（）

答案:對Establishthebehaviormechanismmodelofartificialneuralnetwork,whichisascienceandtechnologytoabstracthumanbrainneuronsfromtheperspectiveofinformationprocessing,andformdifferentnetworkstructuresaccordingtodifferentconnectionmodes,andthensimulatetheactionofneuralnetwork.（）

答案:對ThestructureofANNiscomposedofalargenumberofprocessingunitsin().

答案:parallelThestrengthofsynaptictransmissionisnotunchangeable,itwillshowchangingtransmissionefficiency.（）

答案:對TheerrorfunctionusedinBPneuralnetworkalgorithmis（）.

答案:meansquareerrorfunctionInrecentyears,deeplearninghasshownexcellentperformanceinsolvingmanyproblemssuchasvisualrecognition,speechrecognitionandnaturallanguageprocessing.（）

答案:對Theinputlayerisresponsibleforreceivinginformationfromtheoutsideworldandtransmittingittothenextlayerofneurons.（）

答案:對Hopfieldneuralnetworkiscalled（）

答案:associativememoryInLMSlearningalgorithm,theweightvariationofADALINEneuralnetworkisinverselyproportionaltotheinputandoutputerrorsofthenetwork.（）

答案:錯Thelearningstrategyofperceptronistoselectthemodelparameterswandbthatminimizethelossfunction.（）

答案:minimizeThedescriptionoftheparameter"validator"ofthewx.TextCtrlclassis:the（）.

答案:validatorofcontrolWhenthewindoweventoccurs,themaineventloopwillrespondandassigntheappropriateeventhandlertothewindowevent.（）

答案:對Thedescriptionoftheparameter"defaultDir"ofclasswx.FileDialogis:（）.

答案:defaultpathFromtheuser'spointofview,thewxPythonprogramisidleforalargepartofthetime,butwhentheuserortheinternalactionofthesystemcausestheevent,andthentheeventwilldrivethewxPythonprogramtoproducethecorrespondingaction.（）

答案:對InthedesignofartificialneuralnetworksoftwarebasedonwxPython,creatingGUImeansbuildingaframeworkinwhichvariouscontrolscanbeaddedtocompletethedesignofsoftwarefunctions.（）

答案:對Thedescriptionoftheproperty"tooltipstring"isthepromptthatappearswhenthemouseisovertheobject.（）

答案:對Itisnecessarytodeterminethestructureandparametersoftheneuralnetwork,includingthenumberofhiddenlayers,thenumberofneuronsinthehiddenlayerandthetrainingfunction.（）

答案:對Themenupropertybarhas"label"and"tag".Thelabelisequivalenttothetagvalueofthemenuitem,andthetagisthenameofthemenudisplay.（）

答案:錯Inlarge-scalesystemsoftwaredesign,weneedtoconsiderthelogicalstructureandphysicalstructureofsoftwarearchitecture.（）

答案:對Thedescriptionoftheproperty"string"is:thetextdisplayedontheobject.（）

答案:對The"netprod"inthenetworkinputmodulecanbeusedfor（）.

答案:dotmultiplicationordotdivisionTheneuralnetworktoolboxcontains（）modulelibraries.

答案:fiveThe"dotrod"intheweightsettingmoduleisanormaldotproductweightfunction.（）

答案:錯Themathematicalmodelofsingleneuronisy=f(wx+b).（）

答案:對Theneuronmodelcanbedividedintothreeparts:inputmodule,transferfunctionandoutputmodule.（）

答案:對SOFMneuralnetworksaredifferentfromotherartificialneuralnetworksinthattheyadoptcompetitivelearningratherthanbackwardpropagationerrorcorrectionlearningmethodsimilartogradientdescent,andinasense,theyuseneighborhoodfunctionstopreservetopologicalpropertiesofinputspace.（）

答案:對ThecorelayerofSOFMneuralnetworkis（）.

答案:competitionlayerForSOFMneuralnetwork,thecompetitivetransferfunction(CTF)responseis0forthewinningneurons,and1forotherneurons.（）

答案:錯Whentheinputpatterntothenetworkdoesnotbelongtoanypatterninthenetworktrainingsamples,SOFMneuralnetworkcanonlyclassifyitintotheclosestmode.（）

答案:對Inordertodividetheinputpatternsintoseveralclasses,thedistancebetweeninputpatternvectorsshouldbemeasuredaccordingtothesimilarity.（）areusuallyused.

答案:EuclideandistancemethodandcosinemethodThemostbasicpropertyofAdaBoostisthatitreducesthetrainingerrorcontinuouslyinthelearningprocess,thatis,theclassificationerrorrateonthetrainingdatasetuntileachweakclassifieriscombinedintothefinalidealclassifier.（）

答案:對Boostingalgorithmisthegeneralnameofaclassofalgorithms.Theircommongroundistoconstructastrongclassifierbyusingagroupofweakclassifiers.Weakclassifiermainlyreferstotheclassifierwhosepredictionaccuracyisnothighandfarbelowtheidealclassificationeffect.Strongclassifiermainlyreferstotheclassifierwithhighpredictionaccuracy.（）

答案:對ThemainpurposeofaddingregularizationtermintotheformulaofcalculatingstrongclassifieristopreventtheoverfittingofAdaBoostalgorithm,whichisusuallycalledstepsizeinalgorithm.（）

答案:對Amongthemanyimprovedboostingalgorithms,themostsuccessfuloneistheAdaBoost(adaptiveboosting)algorithmproposedbyYoavFreundofUniversityofCaliforniaSanDiegoandRobertSchapireofPrincetonUniversityin1996.（）

答案:對ThelossfunctionofAdaBoostalgorithmis（）.

答案:exponentialfunctionThecharacteristicofElmanneuralnetworkisthattheoutputofthehiddenlayerisdelayedandstoredbythefeedbacklayer,andthefeedbackisconnectedtotheinputofthehiddenlayer,whichhasthefunctionofinformationstorage.（）

答案:對Thefeedbacklayerisusedtomemorizetheoutputvalueoftheprevioustimeofthehiddenlayerunitandreturnittotheinput.Therefore,Elmanneuralnetworkhasdynamicmemoryfunction.（）

答案:對InElmannetwork,thetransferfunctionoffeedbacklayerisnonlinearfunction,andthetransferfunctionofoutputlayerislinearfunction.（）

答案:對TheneuronsinthehiddenlayerofElmannetworkadoptthetangentS-typetransferfunction,whiletheoutputlayeradoptsthelineartransferfunction.Ifthereareenoughneuronsinthefeedbacklayer,thecombinationofthesetransferfunctionscanmakeElmanneuralnetworkapproachanyfunctionwitharbitraryprecisioninfinitetime.（）

答案:對Elmanneuralnetworkisakindofdynamicrecurrentnetwork,whichcanbedividedintofullfeedbackandpartialfeedback.Inthepartialrecurrentnetwork,thefeedforwardconnectionweightcanbemodified,andthefeedbackconnectioniscomposedofagroupoffeedbackunits,andtheconnectionweightcannotbemodified.（）

答案:對Intheartificialneuralnetwork,thequalityofmodelingwilldirectlyaffecttheperformanceofthegenerativemodel,butasmallamountofpriorknowledgeisneededfortheactualcasemodeling.（）

答案:錯AGANmainlyincludesageneratorGandadiscriminatorD.（）

答案:對TheessenceoftheoptimizationprocessofDandGistofindthe（）.

答案:minimaxBecausethegenerativeadversarialnetworkdoesnotneedtodistinguishthelowerboundandapproximateinference,itavoidsthepartitionfunctioncalculationproblemcausedbythetraditionalrepeatedapplicationofMarkovchainlearningmechanism,andimprovesthenetworkefficiency.（）

答案:對Fromtheperspectiveofartificialintelligence,GANusesneuralnetworktoguideneuralnetwork,andtheideaisverystrange.（）

答案:對InAlexNet,thereare650000neuronswithmorethan600000parametersdistributedinfiveconvolutionlayersandthreefullyconnectedlayersandSoftmaxlayerswith1000categories.（）

答案:錯ComparedwithGPU,CPUhashigherprocessingspeed,andhassignificantadvantagesinprocessingrepetitivetasks.（）

答案:錯VGGNetiscomposedoftwoparts:theconvolutionlayerandthefullconnectionlayer,whichcanberegardedasthedeepenedversionofAlexNet.（）

答案:對Atpresent,DCNNhasbecomeoneofthecorealgorithmsinthefieldofimagerecognition,butitisunstablewhenthereisasmallamountoflearningdata.（）

答案:錯Inthefieldoftargetdetectionandclassification,thetaskofthelastlayerofneuralnetworkistoclassify.（）

答案:對Atpresent,researchershavesuccessfullyappliedHopfieldneuralnetworktosolvethetravelingsalesmanproblem(TSP),whichisthemostrepresentativeofoptimizationcombinatorialproblems.（）

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

答案:對Hopfieldneuralnetworkisakindofneuralnetworkwhichcombinesstoragesystemandbinarysystem.Itnotonlyprovidesamodeltosimulatehumanmemory,butalsoguaranteestheconvergenceto（）.

答案:localminimumIn1982,AmericanscientistJohnJosephHopfieldputforwardakindoffeedbackneuralnetwork"Hopfieldneuralnetwork"inhispaperNeuralNetworksandPhysicalSystemswithEmergentCollectiveComputationalAbilities.（）

答案:對Undertheexcitationofinputx,DHNNentersadynamicchangeprocess,untilthestateofeachneuronisnolongerchanged,itreachesastablestate.Thisprocessisequivalenttotheprocessofnetworklearningandmemory,andthefinaloutputofthenetworkisthevalueofeachneuroninthestablestate.（）

答案:對Inpracticalapplication,theinverseofthecorrelationmatrixandthecorrelationcoefficientarenoteasytoobtain,sotheapproximatesteepestdescentmethodisneededinthealgorithmdesign.Thecoreideaisthattheactualmeansquareerrorofthenetworkisreplacedbythemeansquareerrorofthek-thiteration.（）

答案:對Thealgorithmusedinsingle-layerADALINEnetworkisLMSalgorithm,whichissimilartothealgorithmofperceptron,andalsobelongstosupervisedlearningalgorithm.（）

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

答案:對WhentherearemultipleADALINEinthenetwork,theadaptivelinearneuralnetworkisalsocalledMadalinewhichmeansmanyAdalineneuralnetworks.（）

答案:對ADALINEneuralnetworkhassimplestructureandmulti-layerstructure.Itisflexibleinpracticalapplicationandwidelyusedinsignalprocessing,systemidentification,patternrecognitionandintelligentcontrol.（）

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

答案:對RBFneuralnetworkisanovelandeffectivefeedforwardneuralnetwork,whichhasthebestlocalapproximationandglobaloptimalperformance.（）

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

答案:對ForthelearningalgorithmofRBFneuralnetwork,thekeyproblemistodeterminethecenterparametersoftheoutputlayernodereasonably.（）

答案:錯ThemethodofselectingthecenterofRBFneuralnetworkbyself-organizinglearningistoselectthecenterofRBFneuralnetworkbyk-meansclusteringmethod,whichbelongstosupervisedlearningmethod.（）

答案:錯InthestandardBPneuralnetworkalgorithmandmomentumBPalgorithm,thelearningrateisaconstantthatremainsconstantthroughoutthetrainingprocess,andtheperformanceofthelearningalgorithmisverysensitivetotheselectionofthelearningrate.（）

答案:對BPneuralnetworkhasbecomeoneofthemostrepresentativealgorithmsinthefieldofartificialintelligence.Ithasbeenwidelyusedinsignalprocessing,patternrecognition,machinecontrol(expertsystem,datacompression)andotherfields.（）

答案:對L-Malgorithmismainlyproposedforsuperlargescaleneuralnetwork,anditisveryeffectiveinpracticalapplication.（）

答案:錯In1974,PaulWerbosofthenaturalsciencefoundationoftheUnitedStatesfirstproposedtheuseoferrorbackpropagationalgorithmtotrainartificialneuralnetworksinhisdoctoraldissertationofHarvardUniversity,anddeeplyanalyzedthepossibilityofapplyingittoneuralnetworks,effectivelysolvingtheXORloopproblemthatsinglesensorcannothandle.（）

答案:對TheoutputofBPneuralnetworkis（）ofneuralnetwork.

答案:theoutputofthelastlayerWhentheperceptronislearning,eachsamplewillbeinputintotheneuronasastimulus.Theinputsignalisthefeatureofeachsample,andtheexpectedoutputisthecategoryofthesample.Whentheoutputisdifferentfromthecategory,wecanadjustthesynapticweightandbiasvalueuntiltheoutputofeachsampleisthesameasthecategory.（）

答案:對Theperceptronlearningalgorithmisdrivenbymisclassification,sothestochasticgradientdescentmethodisusedtooptimizethelossfunction.（）

答案:misclassificationThebasicideaofperceptronlearningalgorithmistoinputsamplesintothenetworkstepbystep,andadjusttheweightmatrixofthenetworkaccordingtothedifferencebetweentheoutputresultandtheidealoutput,thatistosolvetheoptimizationproblemoflossfunctionL(w,b).（）

答案:對

答案:對Perceptronisasingle-layerneuralnetwork,orneuron,whichisthesmallestunitofneuralnetwork.（）

答案:對LogarithmicS-typetransferfunction,namelySigmoidfunction,isalsocalledS-shapedgrowthcurveinbiology.（）

答案:對Thefunctionoftransferfunctioninneuronsistogetanewmappingoutputofsummeraccordingtothespecifiedfunctionrelationship,andthencompletes