算法类外文资料翻译

《算法类外文资料翻译》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、Q-LearningByExamplesInthistutorial,youwilldiscoverstepbystephowanagentlearnsthroughtrainingwithoutteacher(unsupervised)inunknownenvironment.YouwillfindoutpartofreinforcementlearningalgorithmcalledQ-learning.Reinforcementlearningalgorithmhasbeenwidelyusedformanyapplicationssuchasroboti

2、cs,multiagentsystem,game,andetc.Insteadoflearningthetheoryofreinforcementthatyoucanreaditfrommanybooksandotherwebsites(seeResourcesformorereferences),inthistutorialwillintroducetheconceptthroughsimplebutcomprehensivenumericalexample.YoumayalsodownloadtheMatlabcodeorMSExcelSpreadsheetf

3、orfree.Supposewehave5roomsinabuildingconnectedbycertaindoorsasshowninthefigurebelow.WegivenametoeachroomAtoE.Wecanconsideroutsideofthebuildingasonebigroomtocoverthebuilding,andnameitasF.NoticethattherearetwodoorsleadtothebuildingfromF,thatisthroughroomBandroomE.Wecanrepresenttheroomsb

4、ygraph,eachroomasavertex(ornode)andeachdoorasanedge(orlink).RefertomyothertutorialonGraphifyouarenotsureaboutwhatisGraph.10Wewanttosetthetargetroom.Ifweputanagentinanyroom,wewanttheagenttogooutsidethebuilding.Inotherword,thegoalroomisthenodeF.Tosetthiskindofgoal,weintroducegiveakindof

5、rewardvaluetoeachdoor(i.e.edgeofthegraph).Thedoorsthatleadimmediatelytothegoalhaveinstantrewardof100(seediagrambelow,theyhaveredarrows).Otherdoorsthatdonothavedirectconnectiontothetargetroomhavezeroreward.Becausethedooristwoway(fromAcangotoEandfromEcangobacktoA),weassigntwoarrowstoeac

6、hroomofthepreviousgraph.Eacharrowcontainsaninstantrewardvalue.ThegraphbecomesstatediagramasshownbelowAdditionalloopwithhighestreward(100)isgiventothegoalroom(FbacktoF)sothatiftheagentarrivesatthegoal,itwillremainthereforever.Thistypeofgoaliscalledabsorbinggoalbecausewhenitreachesthego

7、alstate,itwillstayinthegoalstate.Ladiesandgentlemen,nowisthetimetointroduceoursuperstaragent….Imagineouragentasadumbvirtualrobotthatcanlearnthroughexperience.Theagentcanpassoneroomtoanotherbuthasnoknowledgeoftheenvironment.Itdoesnotknowwhichsequenceofdoorstheagentmustpasstogooutsideth

8、ebuil