欢迎来到天天文库
浏览记录
ID:38593919
大小:1.12 MB
页数:18页
时间:2019-06-15
《哈工大机器学习历年考试》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、1Givethedefinitionsoryourcomprehensionsofthefollowingterms.(12’)1.1TheinductivelearninghypothesisP171.2OverfittingP491.4ConsistentlearnerP1482Givebriefanswerstothefollowingquestions.(15’)2.2Ifthesizeofaversionspaceis,Ingeneralwhatisthesmallestnumberofqueriesmayberequiredbyaconceptlearnerusi
2、ngoptimalquerystrategytoperfectlylearnthetargetconcept?P272.3Ingenaral,decisiontreesrepresentadisjunctionofconjunctionsofconstrainsontheattributevaluesofinstanse,thenwhatexpressiondoesthefollowingdecisiontreecorrespondsto?OutLookHumidityWindSunnyOvercastRainYesHighNormalYesNoStrongYesWeakNo
3、3Givetheexplainationtoinductivebias,andlistinductivebiasofCANDIDATE-ELIMINATIONalgorithm,decisiontreelearning(ID3),BACKPROPAGATIONalgorithm.(10’)4Howtosolveoverfittingindecisiontreeandneuralnetwork?(10’)Solution:lDecisiontree:u及早停止树增长(stopgrowingearlier)u后修剪法(post-pruning)lNeuralNetworku权值衰
4、减(weightdecay)u验证数据集(validationset)5ProvethattheLMSweightupdateruleperformsagradientdescenttominimizethesquarederror.Inparticular,definethesquarederrorEasinthetext.NowcalculatethederivativeofEwithrespecttotheweight,assumingthatisalinearfunctionasdefinedinthetext.Gradientdescentisachievedbyu
5、pdatingeachweightinproportionto.Therefore,youmustshowthattheLMStrainingrulealtersweightsinthisproportionforeachtrainingexampleitencounters.()(8’)Solution:AsVtrain(b)ß(Successor(b))wecangetE==AsmentionedinLMS:WecangetTherefore,gradientdescentisachievementbyupdatingeachweightinproportionto;LM
6、Srulesaltersweightsinthisproportionforeachtrainingexampleitencounters.6Trueorfalse:ifdecisiontreeD2isanelaborationoftreeD1,thenD1ismore-general-thanD2.AssumeD1andD2aredecisiontreesrepresentingarbitrarybooleanfuncions,andthatD2isanelaborationofD1ifID3couldextendD1toD2.Iftruegiveaproof;iffals
7、e,acounterexample.(Definition:Letandbeboolean-valuedfunctionsdefinedover.thenismore_general_than_or_equal_to(written)Ifandonlyifthen)(10’)Thehypothesisisfalse.OnecounterexampleisAXORBwhileifA!=B,trainingexamplesareallpositive,whileifA==B,trainingexamples
此文档下载收益归作者所有