Chapter 5 - Markov Devision Processes v1

Chapter 5 - Markov Devision Processes v1

ID:40709680

大小:4.60 MB

页数:91页

时间:2019-08-06

Chapter 5 - Markov Devision Processes v1_第1页
Chapter 5 - Markov Devision Processes v1_第2页
Chapter 5 - Markov Devision Processes v1_第3页
Chapter 5 - Markov Devision Processes v1_第4页
Chapter 5 - Markov Devision Processes v1_第5页
资源描述:

《Chapter 5 - Markov Devision Processes v1》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、§5MarkovDecisionProcessesCOMP7404Computationalintelligenceandmachinelearning!Dr.DirkSchniedersAcknowledgements:BasedonmaterialsfromCS188@Berkeley&TextbookChapter17SequentialDecisionProblem★Inasequentialdecisionproblemtheagent’sutilitydependsonasequenceofdecision

2、s★Sequentialdecisionproblemsincorporateutilities,uncertaintyandsensing★Optimalbehaviorbalancestherisksandrewardsofactinginanuncertainenvironment2013-2014COMP7404!2Example:GridWorld★Maze-likeproblemspecialexitstateexitstates★Agentlivesinagrid★Wallsblocktheagent’s

3、patha★Unreliableactions★Eachactionachievestheintendedeffectb80%ofthetime★10%ofthetimetheactionmovesthecagentatrightanglestotheintendeddirection1234★IftheagentbumpsintoawallitstaysinthesamesquareExample(IntendtogoNorth):★Theagentreceivesrewardsforeachaction★Small

4、“living”rewards(canbenegative)0.8N★Bigrewardscomeattheend(goodorbad)★Goal:maximizesumofrewards0.1WE0.12013-2014COMP7404!3GridWorld-StochasticmotionDeterministicactionStochasticaction0.10.80.12013-2014COMP7404!4GridWorld-Stochasticmotionreward=0.0reward=-0.1rewar

5、d=-0.2reward=-0.3exitreward=0.4reward=-0.6reward=-0.5reward=-0.42013-2014COMP7404!5Example:GridWorld★Considerthefixedsequence[North,North,East,East,East]a★Canwereachthegoalstateata4withthissequence?b★Whatistheprobability?c★Probabilityofalwaystakingtheintendedact

6、ionis0.851234★Thereisalsoachanceof0.8Naccidentallyreachingthegoalbygoingtheotherwayaroundwithprobability0.14×0.80.1WE0.1★Overallchanceofreaching(4,3)is0.327762013-2014COMP7404!6Quiz★Whichsquarescanbereachedfromc1bytheactionsequence[East,East,East,North,North]and

7、withwhatprobabilities?abc12342013-2014COMP7404!7Transitionmodel★ThetransitionmodelT(s,a,s’)describestheoutcomeofeachactionineachstate★TheoutcomeisstochasticandwewriteP(s’

8、s,a)todenotetheprobabilityofreachingstates’ifactionaisdoneinstates★Weassumethattransitionsa

9、reMarkovian★I.e.,theprobabilityofreachings’fromsdependsonlyonsandnotonthehistoryofearlierstates★TheMarkovpropertyisnamedaftertheRussianmathematicianAndreyMarkov2013-2

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。