Reinforcement Learning_an _introduction.pdf

《Reinforcement Learning_an _introduction.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、BookNext:ContentsContentsReinforcementLearning:AnIntroductionRichardS.SuttonandAndrewG.BartoABradfordBookTheMITPressCambridge,MassachusettsLondon,EnglandInmemoryofA.HarryKlopf●Contents❍Preface❍SeriesForward❍SummaryofNotation●I.TheProblem❍1.Introduction■1.1ReinforcementLearninght

2、tp://www.cs.ualberta.ca/%7Esutton/book/ebook/the-book.html(1di4)22/06/20059.04.27Book■1.2Examples■1.3ElementsofReinforcementLearning■1.4AnExtendedExample:Tic-Tac-Toe■1.5Summary■1.6HistoryofReinforcementLearning■1.7BibliographicalRemarks❍2.EvaluativeFeedback■2.1An-ArmedBanditProb

3、lem■2.2Action-ValueMethods■2.3SoftmaxActionSelection■2.4EvaluationVersusInstruction■2.5IncrementalImplementation■2.6TrackingaNonstationaryProblem■2.7OptimisticInitialValues■2.8ReinforcementComparison■2.9PursuitMethods■2.10AssociativeSearch■2.11Conclusions■2.12BibliographicalandH

4、istoricalRemarks❍3.TheReinforcementLearningProblem■3.1TheAgent-EnvironmentInterface■3.2GoalsandRewards■3.3Returns■3.4UnifiedNotationforEpisodicandContinuingTasks■3.5TheMarkovProperty■3.6MarkovDecisionProcesses■3.7ValueFunctions■3.8OptimalValueFunctions■3.9OptimalityandApproximat

5、ion■3.10Summary■3.11BibliographicalandHistoricalRemarks●II.ElementarySolutionMethods❍4.DynamicProgramming■4.1PolicyEvaluation■4.2PolicyImprovement■4.3PolicyIteration■4.4ValueIteration■4.5AsynchronousDynamicProgramming■4.6GeneralizedPolicyIteration■4.7EfficiencyofDynamicProgrammi

6、nghttp://www.cs.ualberta.ca/%7Esutton/book/ebook/the-book.html(2di4)22/06/20059.04.27Book■4.8Summary■4.9BibliographicalandHistoricalRemarks❍5.MonteCarloMethods■5.1MonteCarloPolicyEvaluation■5.2MonteCarloEstimationofActionValues■5.3MonteCarloControl■5.4On-PolicyMonteCarloControl■

7、5.5EvaluatingOnePolicyWhileFollowingAnother■5.6Off-PolicyMonteCarloControl■5.7IncrementalImplementation■5.8Summary■5.9BibliographicalandHistoricalRemarks❍6.Temporal-DifferenceLearning■6.1TDPrediction■6.2AdvantagesofTDPredictionMethods■6.3OptimalityofTD(0)■6.4Sarsa:On-PolicyTDCon

8、trol■6.5Q-Learning:Off-PolicyTDControl■6.6Actor