Production Matching for Large Learning Systems

Production Matching for Large Learning Systems

ID:39813668

大小:1.13 MB

页数:208页

时间:2019-07-11

Production Matching for Large Learning Systems_第1页
Production Matching for Large Learning Systems_第2页
Production Matching for Large Learning Systems_第3页
Production Matching for Large Learning Systems_第4页
Production Matching for Large Learning Systems_第5页
资源描述:

《Production Matching for Large Learning Systems》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、ProductionMatchingforLargeLearningSystemsRobertB.DoorenbosJanuary31,1995CMU-CS-95-113ComputerScienceDepartmentCarnegieMellonUniversityPittsburgh,PASubmittedinpartialful llmentoftherequirementsforthedegreeofDoctorofPhilosophy.ThesisCommittee:JillFainLehma

2、n,ComputerScienceDepartmentTomMitchell,ComputerScienceDepartmentManuelaVeloso,ComputerScienceDepartmentPaulRosenbloom,UniversityofSouthernCalifornia/InformationSciencesInstitutec1995RobertB.DoorenbosThisresearchwassponsoredinpartbyaNationalScienceFoundat

3、ionGraduateFellowship,andbytheWrightLaboratory,AeronauticalSystemsCenter,AirForceMaterielCommand,USAF,andtheAdvancedResearchProjectsAgency(ARPA)undergrantnumberF33615-93-1-1330.Viewsandconclusionscontainedinthisdocumentarethoseoftheauthorandshouldnotbein

4、terpretedasnecessarilyrepresentingocialpoliciesorendorsements,eitherexpressedorimplied,ofWrightLaboratoryortheUnitedStatesGovernment.TheU.S.GovernmentisauthorizedtoreproduceanddistributereprintsforGovernmentalpurposes,notwithstandinganycopyrightnotation

5、thereon.Keywords:arti cialintelligence,large-scaleproductionsystems,machinelearning,produc-tionmatch,SoarAbstractOneofthecentralresultsofAIresearchinthe1970'swasthattoachievegoodperformance,AIsystemsmusthavelargeamountsofknowledge.Machinelearningtechniqu

6、eshavebeendevelopedtoautomaticallyacquireknowledge,oftenintheformofif-thenrules(productions).Unfortunately,thisoftenleadstoautilityproblem

7、thelearning"endsupcausinganoverallslowdowninthesystem.Thisisbecausethemorerulesasystemhas,thelongerittakestomatcht

8、hemagainstthecurrentsituationinordertodeterminewhichonesareapplicable.Toaddressthisproblem,thisthesisisaimedatenablingthescalingupofthenumberofrulesinproductionsystems.Weexamineadiversesetoftestbedsystems,eachofwhichlearnsatleast100,000rules.Weshowthatwi

9、ththebestexistingmatchalgorithms,thematchcostincreaseslinearlyinthenumberofrulesinthesesystems.Thisisinadequateforlargelearningsystems,becauseitleadstoautilityproblem.Wethenexaminethecausesofthislinearincrease,anddevelopte

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。