会员注册 | 登录 | 微信快捷登录 支付宝快捷登录 QQ登录 微博登录 | 帮助中心 人人文库renrendoc.com美如初恋!
站内搜索 百度文库

热门搜索: 直缝焊接机 矿井提升机 循环球式转向器图纸 机器人手爪发展史 管道机器人dwg 动平衡试验台设计

   首页 人人文库网 > 资源分类 > PDF文档下载

外文资料--Machine learning and games.pdf

  • 资源星级:
  • 资源大小:158.67KB   全文页数:5页
  • 资源格式: PDF        下载权限:注册会员/VIP会员
您还没有登陆,请先登录。登陆后即可下载此文档。
  合作网站登录: 微信快捷登录 支付宝快捷登录   QQ登录   微博登录
友情提示
2:本站资源不支持迅雷下载,请使用浏览器直接下载(不支持QQ浏览器)
3:本站资源下载后的文档和图纸-无水印,预览文档经过压缩,下载后原文更清晰   

外文资料--Machine learning and games.pdf

MachLearn200663211–215DOI10.1007/s109940068919xGUESTEDITORIALMachinelearningandgamesMichaelBowlingJohannesF¨urnkranzThoreGraepelRonMusickPublishedonline10May2006SpringerScienceBusinessMedia,LLC2006ThehistoryoftheinteractionofmachinelearningandcomputergameplayinggoesbacktotheearliestdaysofArtificialIntelligence,whenArthurSamuelworkedonhisfamouscheckerplayingprogram,pioneeringmanymachinelearningandgameplayingtechniquesSamuel,1959,1967.Sincethen,bothfieldshaveadvancedconsiderably,andresearchintheintersectionofthetwocanbefoundregularlyinconferencesintheirrespectivefieldsandingeneralAIconferences.ForsurveysofthefieldwerefertoGinsberg1998,Schaeffer2000,F¨urnkranz2001editedvolumeshavebeencompiledbySchaefferandvandenHerik2002andbyF¨urnkranzandKubat2001.Inrecentyears,thecomputergamesindustryhasdiscoveredAIasanecessaryingredienttomakegamesmoreentertainingandchallengingand,viceversa,AIhasdiscoveredcomputergamesasaninterestingandrewardingapplicationarea.TheindustrysperspectiveiswitnessedbyaplethoraofrecentbooksongentleintroductionstoAItechniquesforgameprogrammersCollins,2002Champanard,2003BourgSeemann,2004Schwab,2004oraseriesofeditedcollectionsofarticlesRabin,2002,2003,2006.AIresearchoncomputergamesbegantofollowdevelopmentsinthegamesindustryearlyon,butsinceJohnLairdskeynoteaddressattheAAAI2000conference,inwhichheadvocatedInteractiveComputerGamesasachallengingandrewardingapplicationareaforAILairdvanLent,2001,numerousworkshopsFuOrkin,2004Ahaetal.,2005,conferences,andspecialissuesofjournalsForbusLaird,2002demonstratethegrowingimportanceofgameplayingapplicationsforArtificialIntelligence.M.Bowlingenvelopebackemailbowlingcs.ualberta.caJ.F¨urnkranzemailfuernkranzinformatik.tudarmstadt.deT.Graepelemailthoregmicrosoft.comR.Musickemailmusickikuni.comSpringer212MachLearn200663211–215Games,whethercreatedforentertainment,simulation,oreducation,providegreatopportunitiesformachinelearning.ThevarietyofpossiblevirtualworldsandthesubsequentMLrelevantproblemsposedfortheagentsinthoseworldsislimitedonlybytheimagination.Furthermore,notonlyisthegamesindustrylargeandgrowinghavingsurpassedthemovieindustryinrevenueafewyearsback,butitisfacedwithatremendousdemandfornoveltythatitstrugglestoprovide.Againstthisbackdrop,machinelearningdrivensuccesseswoulddrawhighprofileattentiontothefield.Surprisinglyhowever,themorecommercialthegametodate,thelessimpactlearninghasmade.ThisisquiteunlikeothergreatmatchesbetweenapplicationanddatadrivenanalyticssuchasdataminingandOLAP.Topicsofparticularimportanceforsuccessfulgameapplicationsincludelearninghowtoplaythegamewell,playermodeling,adaptivity,modelinterpretationandofcourseperformance.TheseneedscanberecastasacallfornewpracticalandtheoreticaltoolstohelpwithlearningtoplaythegameGameworldsprovideexcellenttestbedsforinvestigatingthepotentialtoimproveagentscapabilitiesvialearning.Theenvironmentcanbeconstructedwithvaryingcharacteristics,fromdeterministicanddiscreteasinclassicalboardandcardgamestonondeterministicandcontinuousasinactioncomputergames.Learningalgorithmsforsuchtaskshavebeenstudiedquitethoroughly.ProbablythebestknowninstanceofalearninggameplayingagentistheBackgammonplayingprogramTDGammonTesauro,1995.learningaboutplayersOpponentmodeling,partnermodeling,teammodeling,andmultipleteammodelingarefascinating,interdependentandlargelyunsolvedchallengesthataimatimprovingplaybytryingtodiscoverandexploittheplans,strengths,andweaknessesofaplayersopponentsand/orpartners.OneofthegrandchallengesinthislineofworkaregameslikePoker,whereopponentmodelingiscrucialtoimproveovergametheoreticallyoptimalplayBillingsetal.,2002.behaviorcaptureofplayersCreatingaconvincingavatarbasedonaplayersingamebehaviorisaninterestingandchallengingsupervisedlearningtask.Forexample,inMassiveMultiplayerOnlineRoleplayingGamesMMORGsanavatarthatistrainedtosimulateausersgameplayingbehaviorcouldtakehiscreatorsplaceattimeswhenthehumanplayercannotattendtohisgamecharacter.FirststepsinthisareahavebeenmadeincommercialvideogamessuchasForzaMotorsportXboxwheretheplayercantrainaDrivatarthatlearnstogoaroundthetrackinthestyleoftheplayerbyobservingandlearningfromthedrivingstyleofthatplayerandgeneralizingtonewtracksandcars.modelselectionandstabilityOnlinesettingsleadtowhatiseffectivelytheunsupervisedconstructionofmodelsbysupervisedalgorithms.Methodsforbiasingtheproposedmodelspacewithoutsignificantlossofpredictivepowerarecriticalnotjustforlearningefficiency,butinterpretiveabilityandenduserconfidence.optimizingforadaptivityBuildingopponentsthatcanjustbarelyloseininterestingwaysisjustasimportantforthegameworldascreatingworldclassopponents.Thisrequiresbuildinghighlyadaptivemodelsthatcansubstantivelypersonalizetoadversariesorpartnerswithawiderangeofcompetenceandrapidshiftsinplaystyle.Byintroducingaverydifferentsetofupdateandoptimizationcriteriaforlearners,awealthofnewresearchtargetsarecreated.modelinterpretationWhatsmynextmoveisnottheonlyquerydesiredofmodelsinagame,butitiscertainlytheonewhichgetsthemostattention.Creatingtheillusionofintelligencerequirespaintingapictureofanagentsthinkingprocess.TheabilitytodescribethecurrentstateofamodelandtheprocessofinferenceinthatmodelfromSpringerMachLearn200663211–215213decisiontodecisionenablesqueriesthatprovidethefoundationforahostofsocialactionsinagamesuchaspredictions,contracts,counterfactualassertions,advice,justification,negotiation,anddemagoguery.Thesecanhaveasmuchormoreinfluenceonoutcomesasactualingameactions.performanceResourcerequirementsforupdateandinferencewillalwaysbeofgreatimportance.TheAIdoesnotgetthebulkoftheCPUormemory,andthemachinesdrivingthemarketwillalwaysbeunderpoweredcomparedtotypicaldesktopsatanypointintime.Thisspecialissuecontainsthreearticlesandoneresearchnotethatspanthewiderangeofresearchintheintersectionofgameplayingandmachinelearning.Inthefirstcontribution,AdaptiveGameAIwithDynamicScripting,Sproncketal.tackletheproblemofadaptivitybydynamicallymodifyingtheruleswhichgoverncharacterbehavioringame.Thispaperistargetedatthecommercialgamesindustry,andprovidessomegoodinsightintoproblemsfacedbythecreatorsoftodaysroleplayinggames.Theauthorsproposefourfunctionalandfourcomputationalrequirementsforonlinelearningingames.Theythenproceedtoshowhowdynamicscriptingfitsintothoserequirements,andprovideexperimentalevidenceofthepotentialpromiseofthisapproach.Dynamicscriptingcanbecharacterizedasstochasticoptimization.Theauthorsevaluatedynamicscriptingonboththetaskofprovidingthetoughestopponentpossible,andonthetaskofdifficultyscaling.Gooddifficultyscalingunderpinswhatmakesmostgamesfun,andsolvingthisproblemisoftenverychallengingandthesolutionsarealmostalwaysadhoc.TheauthorspresentexperimentaldatathatcomparesdynamicscriptingtostaticopponentsandthosecontrolledbyQLearningandMonteCarlo.ThetestenvironmentsincludebothsimulatedgamesandanactualcommercialgameNeverwinterNights,andhelptopresentaveryinterestingstudywhichissuretoblazeapathforfurtherinterestingresearch.Thesecondpaper,UniversalParameterOptimizationinGamesBasedonSPSAbySzepesvariandKocsis,considerstheproblemofoptimizingparameterstoimprovetheperformanceofparameterizedpoliciesforgameplay.TheyconsidertheSimultaneousPerturbationStochasticApproximationSPSAmethodintroducedbySpall1992whichisageneralgradientfreeoptimizationmethodthatisapplicabletoawiderangeofoptimizationproblems.TheauthorsdemonstratethatSPSAisapplicabletoawiderangeoftypicaloptimizationproblemsingamesandproposeseveralmethodstoenhancetheperformanceofSPSA.Theseenhancementsincludetheuseofcommonrandomnumbersandantitheticvariables,acombinationwithRPROPandthereuseofsamples.TheapplicationtogamesconsidersthedomainoflearningtoplayOmahaHiLoPokerwiththeirpokerprogramMcRaise.SPSAcombinedwiththeirproposedenhancementsleadstopokerperformancecompetitivewithTDlearning,themethodsosuccessfullyusedbyTesauro1995,forlearningaworldclassevaluationfunctionforBackgammonandstillusedintodaysworldclassbackgammonprogramssuchasJellyFishandSnowie.Thethirdcontribution,LearningtoBidinBridgebyMarkovitchandAmit,addressestheproblemofbiddinginthegameofBridge.WhileresearchinBridgeplayinghaspioneeredMonteCarlosearchalgorithmsfortheplayingphaseofcardgamesandresultedinprogramsofconsiderablestrengthGinsberg,1999,thebiddingphase,inwhichthegoalthesocalledcontractofthesubsequentplayingphaseisdetermined,isstillamajorweaknessofexistingBridgeprograms.ThispaperisaboutanapproachthatsupportsthedifficultbiddingphaseinthegameBridgewithtechniquesfrommachinelearning,inparticularopponentmodelingviathelearningofdecisionnetsandviamodelbasedMonteCarlosamplingtoaddresstheproblemofhiddeninformation.Theevaluationclearlyestablishesthatthesystemimproveswithlearning,anditseemsthatthelevelofplayachievedbythisprogramsurpassesthelevelSpringer214MachLearn200663211–215ofthebiddingmoduleofcurrentstateoftheartprogramsandapproachesthatofanexpertplayer.Finally,SadikovandBratkopresentaresearchnoteonLearningLongtermChessStrategiesfromDatabases.Theyaddresstheproblemofknowledgediscoveryingamedatabases.Formanygamesorsubgamessuchaschessendgames,therearegamedatabasesavailable,whichcontainperfectinformationaboutthegameinthesensethatforeverypossibleposition,thegametheoreticoutcomeisstoredinadatabase.However,althoughthesedatabasescontainallinformationtoallowperfectplay,theyarenotamenabletohumananalysis,andaretypicallynotverywellunderstood.Forexample,chessGrandmasterJohnNunnanalyzedseveralsimplechessendgamedatabasesresultinginaseriesofwidelyacknowledgedendgamebooksNunn,1992,1994b,1995,butreadilyadmittedthathedoesnotyetunderstandallaspectsofthedatabasesheanalyzedNunn,1994a.Thispaperreportsonanattempttomakeheadwaybyautomaticallyconstructingplayingstrategiesfromchessendgamedatabases.Itdescribesamethodforbreakinguptheproblemintodifferentgamephases.Foreachphase,itisthenproposedtolearnaseparateevaluationfunctionvialinearregression.Experimentsinthethekingandrookvs.king,orkingandqueenvs.kingandrookendgamesshowencouragingresults,butalsoillustratethedifficultyoftheproblem.MachinelearninghasbeeninstrumentaltodateinbuildingsomeoftheworldsbestplayersinBackgammonandhasleadtointerestingresultsingameslikeChessandGo.Tomoveintomainstreamcommercialgames,machinelearningresearchhastofacewhatinmanywaysaretheharderproblemsoflosingininterestingways,creatingmoreusefulillusionsofintelligence,hyperfastadaptation,andtakingonpersona.Thearticlesinthisspecialissueprovideaglimpseintodifferentfacetsofalloftheseproblems.ReferencesAha,D.W.,Mu˜nozAvilaH.M.,vanLent,M.Eds.,2005.Reasoning,representation,andlearningincomputergamesProceedingsoftheIJCAIworkshop.Edinburgh,ScotlandNavalResearchLaboratory,NavyCenterforAppliedResearchinArtificialIntelligence.TechnicalReportAIC05127.Billings,D.,Pe˜na,L.,Schaeffer,J.,Szafron,D.2002.Thechallengeofpoker.ArtificialIntelligence,1341–2,201–240,SpecialIssueonGames,ComputersandArtificialIntelligence.Bourg,D.M.,SeemannG.2004.AIforgamedevelopersCreatingintelligentbehavioringames.OReilly.Champanard,A.2003.AIgamedevelopment.NewRidersPublishing.Collins,M.2002.AdvancedAIgamedevelopment.WordwarePublishingInc.Forbus,K.D.,LairdJ.E.2002.GuesteditorsintroductionAIandtheentertainmentindustry.IEEEIntelligentSystems,174,15–16.Fu,D.OrkinJ.Eds.,2004.ChallengesofgameAIProceedingsoftheAAAI04workshop.AAAIPress.TechnicalReportWS0404.F¨urnkranz,J.2001.MachinelearningingamesAsurvey.InF¨urnkranzandKubat,2001,pp.11–59.F¨urnkranz,J.,KubatM.Eds.,2001.Machinesthatlearntoplaygames.Huntington,NYNovaSciencePublishers.Ginsberg,M.L.1998.Computers,gamesandtherealworld.ScientificAmericanPresents94.SpecialIssueonExploringIntelligence.Ginsberg,M.L.1999.GIBStepstowardanexpertlevelbridgeplayingprogram.InProceedingsoftheInternationalJointConferenceonArtificialIntelligenceIJCAI99,pp.584–589,Stockholm,Sweden.Laird,J.E.,vanLent,M.2001.HumanlevelAIskillerapplicationInteractivecomputergames.AIMagazine,222,15–26.Nunn,J.1992.Secretsofrookendings.Batsford.Nunn,J.1994a.Extractinginformationfromendgamedatabases.InH.J.vandenHerik,I.S.Herschberg,andJ.W.H.M.UiterwijkEds.,Advancesincomputerchess7,pp.19–34.Maastricht,TheNetherlandsUniversityofLimburg.Nunn,J.1994b.Secretsofpawnlessendings.Batsford.Springer

注意事项

本文(外文资料--Machine learning and games.pdf)为本站会员(英文资料库)主动上传,人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。 若此文所含内容侵犯了您的版权或隐私,请立即通知人人文库网([email protected]),我们立即给予删除!

温馨提示:如果因为网速或其他原因下载失败请重新下载,重复下载不扣分。

copyright@ 2015-2017 人人文库网网站版权所有
苏ICP备12009002号-5