购买

¥ 20.0

加入VIP
  • 专属下载特权
  • 现金文档折扣购买
  • VIP免费专区
  • 千万文档免费下载

上传资料

关闭

关闭

关闭

封号提示

内容

首页 Introduction to ILP

Introduction to ILP.ppt

Introduction to ILP

精品课件库
2019-06-15 0人阅读 举报 0 0 暂无简介

简介:本文档为《Introduction to ILPppt》,可适用于综合领域

IntroductiontoILPILP=InductiveLogicProgramming=machinelearninglogicprogramming=learningwithlogicIntroducedbyMuggletonin(Machine)LearningTheprocessbywhichrelativelypermanentchangesoccurinbehavioralpotentialasaresultofexperience(Anderson)Learningisconstructingormodifyingrepresentationsofwhatisbeingexperienced(Michalski)AcomputerprogramissaidtolearnfromexperienceEwithrespecttosomeclassoftasksTandperformancemeasureP,ifitsperformanceattasksinT,asmeasuredbyP,improveswithexperienceE(Mitchell)MachineLearningTechniquesDecisiontreelearningConceptualclusteringCasebasedlearningReinforcementlearningNeuralnetworksGeneticalgorithmsand…InductiveLogicProgrammingWhyILPStructureddataSeedexampleofEastWesttrains(Michalski)WhatmakesatraintogoeastwardWhyILP–StructureddataMutagenicityofchemicalmolecules(King,Srinivasan,Muggleton,Sternberg,)WhatmakesamoleculetobemutagenicWhyILP–multiplerelationsThisisrelatedtostructureddatahascarcarpropertiesTrainCartctctctctc……CarLengthShapeAxesRoof…cshortrectanglenone…clongrectanglenone…cshortrectanglepeaked…clongrectanglenone…cshortrectangleflat…………………WhyILP–multiplerelationsGenealogyexample:Givenknownrelations…father(Old,Young)andmother(Old,Young)male(Somebody)andfemale(Somebody)…learnnewrelationsparent(X,Y):father(X,Y)parent(X,Y):mother(X,Y)brother(X,Y):male(X),father(Z,X),father(Z,Y)MostMLtechniquescan’tusemorethanrelationeg:decisiontrees,neuralnetworks,…WhyILP–logicalfoundationProlog=ProgrammingwithLogicisusedtorepresent:Backgroundknowledge(ofthedomain):factsExamples(oftherelationtobelearned):factsTheories(asaresultoflearning):rulesSupportsformsoflogicalreasoningDeductionInductionPrologdefinitionsVariables:X,Y,Something,SomebodyTerms:arthur,,,,Predicates:father,femaleFacts:father(christopher,victoria)female(victoria)Rules:parent(X,Y):father(X,Y)Logicalreasoning:deductionFromrulestofacts…BT|Emother(penelope,victoria)mother(penelope,arthur)father(christopher,victoria)father(christopher,arthur)parent(X,Y):father(X,Y)parent(X,Y):mother(X,Y)parent(penelope,victoria)parent(penelope,arthur)parent(christopher,victoria)parent(christopher,arthur)Logicalreasoning:inductionFromfactstorules…BE|Tmother(penelope,victoria)mother(penelope,arthur)father(christopher,victoria)father(christopher,arthur)parent(X,Y):father(X,Y)parent(X,Y):mother(X,Y)parent(penelope,victoria)parent(penelope,arthur)parent(christopher,victoria)parent(christopher,arthur)InductionofaclassifierorConceptLearningMoststudiedtaskinMachineLearningGiven:backgroundknowledgeBasetoftrainingexamplesEaclassificationcCforeachexampleeFind:atheoryT(orhypothesis)suchthatBT|c(e),foralleEInductionofaclassifier:exampleExampleofEastWesttrainsB:relationshascarandcarproperties(length,roof,shape,etc)ex:hascar(t,c),shape(c,bucket)E:thetrainsttotC:east,westWhyILPStructureddataSeedexampleofEastWesttrains(Michalski)WhatmakesatraintogoeastwardInductionofaclassifier:exampleExampleofEastWesttrainsB:relationshascarandcarproperties(length,roof,shape,etc)ex:hascar(t,c)E:thetrainsttotC:east,westPossibleT:east(T):hascar(T,C),length(C,short),roof(C,)Inductionofaclassifier:exampleExampleofmutagenicityB:relationsatomandbondex:atom(mol,atom,c,)bond(mol,atom,atom,)E:moleculeswithknownclassificationC:activeandnonactivewrtmutagenicityPossibleT:active(Mol):atom(Mol,A,c,),atom(Mol,B,c,),bond(Mol,A,B,)ccLearningassearchGiven:BackgroundknowledgeBTheoryDescriptionLanguageTPositivesexamplesP(class)NegativeexamplesN(class)Acoveringrelationcovers(B,T,e)Find:atheorythatcoversallpositiveexamples(completeness)nonegativeexamples(consistency)LearningassearchCoveringrelationinILPcovers(B,T,e)BT|eAtheoryisasetofrulesEachruleissearchedseparately(efficiency)Arulemustbeconsistent(covernonegatives),butnotnecessarycompleteSeparateandconquerstrategyRemovefromPtheexamplesalreadycoveredSpaceexplorationStrategyRandomwalkRedundancy,incompletenessofthesearchSystematicaccordingtosomeorderingBettercontrol=>noredundancy,completenessTheorderingmaybeusedtoguidethesearchtowardsbetterrulesWhatkindoforderingGeneralityorderingRuleismoregeneralthanrule=>RulecoversmoreexamplesthanruleIfaruleisconsistent(coversnonegatives)theneveryspecialisationofitisconsistenttooIfaruleiscomplete(coversallpositives)theneverygeneralisationofitiscompletetooMeanstoprunethesearchspacekindsofmoves:specialisationandgeneralisationCommonILPordering:θsubsumptionGeneralityorderingparent(X,Y):parent(X,Y):female(X)parent(X,Y):father(X,Y)parent(X,Y):female(X),mother(X,Y)parent(X,Y):female(X),father(X,Y)consistentrulespecialisationSearchbiases“Biasreferstoanycriterionforchoosingonegeneralizationoveranotherotherthanstrictconsistencywiththeobservedtraininginstances”(Mitchell)Restrictthesearchspace(efficiency)Guidethesearch(givendomainknowledge)DifferentkindsofbiasLanguagebiasSearchbiasStrategybiasLanguagebiasChoiceofpredicates:roof(C,flat)roof(C)flat(C)Typesofpredicates:east(T):roof(T),roof(C,)Modesofpredicates:east(T):roof(C,flat)east(T):hascar(T,C),roof(C,flat)DiscretizationofnumericalvaluesSearchbiasThemovesdirectioninthesearchspaceTopdownstart:theemptyrule(c(X):)moves:specialisationsBottomupstart:thebottomclause(~c(X):B)moves:generalisationsBidirectionalStrategybiasHeuristicsearchforabestruleHillclimbing:KeeponlyoneruleefficientbutcanmissglobalmaximumBeamsearch:alsokeepkrulesforbacktrackinglessgreedyBestfirstsearch:keepallrulesmorecostlybutcompletesearchAgenericILPalgorithmprocedureILP(Examples)Initialize(Rules,Examples)repeatR=Select(Rules,Examples)Rs=Refine(R,Examples)Rules=Reduce(RulesRs,Examples)untilStoppingCriterion(Rules,Examples)return(Rules)AgenericILPalgorithmInitialize(Rules,Examples):initializeasetoftheoriesasthesearchstartingpointsSelect(Rules,Examples):selectthemostpromisingcandidateruleRRefine(R,Examples):returnstheneighboursofR(usingspecialisationorgeneralisation)Reduce(Rules,Examples):discardunpromisingtheories(allbutoneinhillclimbing,noneinbestfirstsearch)ILPnet–wwwcsbrisacuk~ILPnetNetworkofExcellenceinILPinEuropeuniversitiesandresearchinstitutesEducationalmaterialsPublicationsEvents(conferences,summerschools,…)DescriptionofILPsystemsApplicationsILPsystemsFOIL(QuinlanandCameronJones):topdownhillclimbingsearchProgol(Muggleton,):topdownbestfirstsearchwithbottomclauseGolem(MuggletonandFeng):bottomuphillclimbingsearchLINUS(LavracandDzeroski):propositionalisationAleph(~Progol),Tilde(relationaldecisiontrees),…ILPapplicationsLifesciencesmutagenecity,predictingtoxicologyproteinstructurefoldingNaturallanguageprocessingenglishverbpasttensedocumentanalysisandclassificationEngineeringfiniteelementmeshdesignEnvironmentalsciencesbiodegradabilityofchemicalcompoundsTheendAfewbooksonILP…JLloydLogicforlearning:learningcomprehensibletheoriesfromstructureddataSDzeroskiandNLavrac,editorsRelationalDataMiningSeptemberLDeRaedt,editorAdvancesinInductiveLogicProgrammingNLavracandSDzeroskiInductiveLogicProgramming:TechniquesandApplications

VIP尊享8折文档

用户评价(0)

关闭

新课改视野下建构高中语文教学实验成果报告(32KB)

抱歉,积分不足下载失败,请稍后再试!

提示

试读已结束,如需要继续阅读或者下载,敬请购买!

文档小程序码

使用微信“扫一扫”扫码寻找文档

1

打开微信

2

扫描小程序码

3

发布寻找信息

4

等待寻找结果

我知道了
评分:

/32

Introduction to ILP

¥20.0

会员价¥16.0

VIP

在线
客服

免费
邮箱

爱问共享资料服务号

扫描关注领取更多福利