资源描述：
《01 - Experimental Goal Regression. A Method for Learning Problem-Solving Heuristics.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、MachineLearning1:249-286,1986©1986KluwerAcademicPublishers,Boston-ManufacturedinTheNetherlandsExperimentalGoalRegression:AMethodforLearningProblem-SolvingHeuristicsBRUCEW.PORTER(PORTER@UTEXAS)ComputerSciencesDepartment,UniversityofTexasatAustin,Austin,
2、TX78712,U.S.A.DENNISF.KIBLER(KIBLER@CIP.UCI.EDU)IrvineComputationalIntelligenceProject,DepartmentofInformationandComputerScience,UniversityofCalifornia,Irvine,CA92717,U.S.A.(ReceivedDecember30,1985)Keywords:goalregression,problemsolvingheuristics,examp
3、legenerationAbstract.Thisresearchexaminestheprocessoflearningproblemsolvingwithminimalrequirementsforaprioriknowledgeandteacherinvolvement.Experienceindicatesthatknowledgeabouttheproblemsolvingtaskcanbeusedtoimproveproblemsolvingperformance.Thisresearc
4、haddressestheissuesofwhatknowledgeisuseful,howitisappliedduringproblemsolving,andhowitcanbeacquired.Foreachoperatorusedintheproblemsolvingdomain,knowledgeisincrementallylearnedconcerningwhyitisuseful,whenitisapplicable,andwhattransformationitperforms.T
5、hemethodofexperimentalgoalregressionisintroducedforimprovingthelearningratebyapproximatingtheresultsofanalyticlearning.Theideasareformalizedinanalgorithmforlearningandproblemsolvinganddemonstratedwithexamplesfromthedomainsofsimultaneouslinearequationsa
6、ndsymbolicintegration.1.IntroductionAlthoughtheearliestsuccessfulresearchinAIwasonproblemsolving(NewellandSimon,1961)andlearning(Samuel,1959)thesymbioticrelationshipbetweenthetwoareaswasnotexploiteduntilrecently(Anzai,1978;Brazdil,1978;Neves,1978).This
7、paperpresentsalearningmethodforautomatingtheacquisitionofknowledgeforproblemsolvingwhichrequiresfewinstances,minimalteacherinvolvement,andon-lyaweaktheoryofthedomain.Amajorissueinthisresearchistheintegrationoflearningandproblemsolving.Thisintegrationpe
8、rmitsthelearnertousetheproblemsolverasanevaluatorandreducesboththeroleoftheteacherandtherequirementforaprioriknowledge.250B.W.PORTERANDD.F.KIBLERAssumptionofahighlycooperativeteacherorextensiveaprioriknowledgelimitstheapplicabilityofmac
