资源描述：
《02 - A General Framework for Induction and a Study of Selective Induction》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、MachineLearning1:177-226,1986©1986KluwerAcademicPublishers,Boston-ManufacturedinTheNetherlandsAGeneralFrameworkforInductionandaStudyofSelectiveInductionLARRYRENDELL(RENDELL@B.CS.UIUC.EDU)DepartmentofComputerScience,UniversityofIllinoisatUrbana-Champaign,Urban
2、a,IL61801,U.S.A.(ReceivedDecember15,1985)Keywords:induction,uncertainandincrementallearning,conceptformationAbstract.Thispaperhastwomajorparts.Thefirstisanextensiveanalysisoftheproblemofinduction,andthesecondpartisadetailedstudyofselectiveinduction.Throughout
3、thepaperweintegrateanumberofnotions,mainlyfromartificialintelligence,butalsofrompatternrecognitionandcognitivepsychology.Theresultisasyntheticviewwhichexploitsuncertainty,task-guidance,andbiasessuchaslanguagerestriction.Someofthemainthemesandcontributionsarea
4、sfollows.(1)Practicalinductionisreallyaproblemofefficacyandefficiency(power).(2)Searchinaspaceofhypotheticalconceptsisgovernedbyacredibilityfunctionwhichcombinesvariousknowledgesourcesinasinglesubjectiveprobabilityorbeliefmeasureu.(3)Theamountofknowledgesuppl
5、iedbyvarioussourcescanoftenbequantified;thesesourcesincludevariousbiasesandthelearningsystemitself.(4)Inductionisequivalenttodiscoveryofautilityfunctionu,whichcapturesthepurposeorgoalofinduction.(5)Thedifficultyofinductionmaybecharacterizedbytheformofu.Smooth
6、orcoherentfunctionsmeanselectiveinduction,whichhashadthemostattentioninmachinelearning.(6)Systemsforselectiveinductionaremoresimilarthancommonlyunderstood.Byjuxtaposingthemwecandiscoversimilaritiesandimprovements.(7)Ouranalysissuggestsanumberofincipientprinci
7、plesforpowerfulinduction.0.Introduction0.1TheprevalenceandnatureofinductionInductionisanimportantbutcomplexproblemwhichhasbeenextensivelystudiedinpsychology,philosophy,patternrecognitionandartificialintelligence.Increasinginteractionamongthesefourfieldshaspro
8、motednewperspectives,andredefinedkernalsfromeachdisciplinemaysoonprovideamorecompleteunderstandingofinduction,alongwithimprovedmechanization.Thispaperpresentssomerecent178LARRYRENDELLfind
