第1題ThefirsttimeAlphaGodefeatedtheworldGochampionin:A2012B2014C2016D2017第2題TheterminologyusedinAItodescribeadecisionisbasedonmaximallyachievingpre-definedgoalsis:ASmartBRational

CIntelligenceDSensible第1題AvideosaidthatMs.Wuseducedaman,sopeopleinferredthatMs.Wuwasabadwoman,andthenthenextdayitwasshownthatthevideowasacheatandpeoplelearnedthatMs.Wuwasagoodgirl.

Theabovereasoningbelongsto:ADeductionBAbductionCInduction第2題Afacetisarelationthathasauniquevalue.

第3題ReificationturnsaN-ary

(N>=3)relationintoanobjectandmakesNbinaryrelations.第4題InaFOLinterpretation,constantsarereferredto:AObjectsBRelationsCFunctionsDTruthRelations第5題Totranslatethefollowingsentence“總有刁民想害朕“intoaFOLsentence,wehave:A?x(刁民(x)∧害朕(x))B?x(刁民(x)→害朕(x))C?x(刁民(x)∧害朕(x))D?x(刁民(x)→害朕(x))第6題IfXisacontradictionandYisasentence,thenX->YisatautologyATrueBFalse第7題Howmanylogicconstantsarethereinfirst-orderlogic?A0B1C2D3EInfinity正確答案：EChapter3Homework第1題SelectwhichofthefollowingsearchmethodsaretrueuninformedsearchAA*SearchBBidirectionalSearchCDepth-limitedSearchDGreedySearch第2題Depth-firstsearch

isanuninformedsearchmethod.第3題Heuristicsearchisasearchstrategythatuseproblem-specificknowledgebeyondthedefinitionoftheproblemitself.

第4題Alpha-betapruningistoaddlargepartsthatareevaluatedbythemini-maxalgorithminitssearchtree.第5題Whichofthefollowingitemsisnot

usedtoformulateaproblem?AAgentfunctionBInitialstateCPathcostDGoaltest第6題Whichofthefollowingisatruestatementaboutgames?

ALocalsearchproblemsareoftenknownasgames.BAdversarialsearchproblemsareoftenknownasgames.CHeuristicsearchproblemsareoftenknownasgames.DClassicalsearchproblemsareoftenknownasgames.第7題Inthefollowingitems,（）

describesasituationinwhichtheinteractingagents'aggregategainsandlossescanbelessthanormorethanzero?AZerosumgameBComputergameCTwo-playergameDNon-zerosumgameChapter4Homework第1題ID3cansuccessfullyavoidoverfitting

inanycase.第2題ThemainreasonsofpruningindecisiontreesNOT

includeAAvoidoverfittingBImprovingtrainingaccuracyCReducetreesizeDImprovingtestaccuracy第3題Inadualperceptrontrainingwiththreelabelsy1,y2,andy3,wemistakenlyclassifiedanewinstancextoy3,whereitshouldbeclassifiedtoy1.Beforeclassifyingx,wehavealpha1=3,alpha2=2,andalpha3=0,thenafterclassifyingx,wehavealpha1=____,alpha2=____,andalpha3=____.正確答案：:4正確答案：:2正確答案：:-1第4題Inagglomerativeclustering,ifwehave10initialinputs,thenaftertwomergingsteps,howmanyclustersdowehave?A1B7C8D9第5題K-NNisaparametricmodelinwhichKistheparameter.

第6題WhichactivationfunctionisNOTagoodchoiceforneuralnetworks?ALinearfunctionBRectifiedlinearCSigmoidDHyperbolictangent第7題Forthe3*3image[3-2

0

2-1

30

41]ifa2*2convolutionkernel

[21-13]isappliedtoit,theoutputshouldbe[____

____

____

____]正確答案：:-1正確答案：:6正確答案：:15正確答案：:0第8題WhichofthefollowingisNOTafeatureofCNN?ACandetectsmallpatternsBRecognizethesameobjectindifferentsizesCInsensitivetorotationDUsesameparameterseverywhereChapter5Homework第1題Whatisanadvantageofthefullyconvolutionalnetwork?AItoutputsalabelfortheinputimageBItmakespredictionsforallpixelsatonce,incomparisontotheslidingwindowmethodCItusesonlyconvolutionandpoolingoperations,justlikeCNNDItworksaswellasCNNforimageclassification第2題Traditionalbooksearchinlibrary(e.g.,TITLEcontains機器學(xué)習(xí)ANDauthorequals周志華)usesABooleanmodelBVectorspacemodelCProbabilisticmodelDDeeplearning第3題WhichnetworktypeisthemostpopularforNLPinrecentyears?AFully-connectedneuralnetworkBConvolutionalneuralnetworkCRecurrent/recursiveneuralnetworkDTransformernetwork第4題Whatisthemostsignificantadvantageofdistributedrepresentation(embedding)inNLP?AOvercomessemanticgapbymatchingsimilartermsBSimplebinaryrepresentationCDoesnotrequireexpensivetrainingDMakesindependenceassumption第5題Whichstatementistrueforvectorspacemodel?AAlltermsmustbepresenttoberetrievedBIthasatermweightingschemeCResultscanberankedbysimilarityDCanmatchsynonyms正確答案：BC第6題WhichofthefollowingisNOTaninformationretrievalapplication?ALookupabookinthelibrarycatalogBSearchforvideoonYoukuwebsiteCMachinetranslationDFindasimilarimagewithBaidusearch第7題WhichstatementisNOTtrue?AForimages,CNNisthemostpopulardeeplearningmodelBCNNsaregettingdeeperforbetterperformanceCCNNmaybefooledifitsparametersareknownDCNNcanonlyshrinkima