资源描述：
《00-Editorial.OnMachineLearning.pdf》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、MachineLearning1:5-10,1986©1986KluwerAcademicPublishers,Boston-ManufacturedinTheNetherlandsEditorial:OnMachineLearningThecentralroleoflearningAlthoughresearchersinartificialintelligenceandpsychologyhavelongrecognizedtheimportanceoflearning,thistopichasnotalwaysbeenthe
2、centralfocusofthesefields.InthefirstyearsofAI,considerableattentionwasgiventolearningissues,butaspatternrecognitionandAIdevelopedseparateidentities,learningresearchbecameassociatedwiththeformerwhilethelatterconcentratedonproblemsofrepresentationandperformance.Asimilar
3、phenomenonoccurredinpsychology.Thebehavioristparadigmwasalmostexclusivelyconcernedwithlearningphenomena,butasinformationprocessingpsychologygainedinpopularity,psychologiststurnedtheirsightstowardsmemoryandperformancephenomenaandallbutabandonedeffortstoexplainthelearni
4、ngprocess.However,thepastfiveyearshaveseenaresurgenceofinterestinlearningwithinbothartificialintelligenceandcognitivepsychology.Thishasresultedpartlyfromdissatisfactionwithpureperformancemodelsofintelligence.Oneofthemajorin-sightsofbothfieldshasbeenthat,exceptinthesim
5、plestdomains,intelligentbehaviorrequiressignificantknowledgeofthosedomains.AlthoughthisinsighthasledtosuccessfulappliedAIsystemsandtoaccuratepsychologicalmodelsofdomain-specificperformance,ithasnotledtosystemsortheorieshavinganygreatdegreeofgenerality.Byrefocusingthei
6、reffortsonlearning,manyresearchershopetodiscovermoregeneralprinciplesofintelligence.Inthecaseofpsychology,suchprin-cipleswouldleadtomoreencompassingtheoriesofhumanbehaviorthatmovebeyondparticulardomains.InthecaseofappliedAI,generallearningmethodsmightletoneautomatethe
7、constructionofknowledge-intensivesystems,savingman-yearsofeffortforeachapplicationarea.Yetdissatisfactionwithperformancemodelsisnotsufficienttoaccountfortheex-plosionofresearchoncomputationalapproachestolearning.Onemustalsocredittheadvancesmadeonrepresentationalandper
8、formanceissueswithinthetwofieldsoverthepasttwodecades.Sinceanylearningsystemmustincorporaterepresenta-tionalandperformanceco
