Artificial Intelligence_ A Modern Approach, 4th Edition.pdf

Artificial Intelligence_ A Modern Approach, 4th Edition.pdf

《Artificial Intelligence_ A Modern Approach, 4th Edition.pdf》由会员分享，可在线阅读，更多相关《Artificial Intelligence_ A Modern Approach, 4th Edition.pdf（2146页珍藏版）》请在冰点文库上搜索。

ArtificialIntelligenceAModernApproachFourthEditionPEARSONSERIESINARTIFICIALINTELLIGENCEStuartRussellandPeterNorvig,EditorsArtificialIntelligenceAModernApproachFourthEditionStuartJ.RussellandPeterNorvigContributingwriters:

Ming-WeiChangJacobDevlinAncaDraganDavidForsythIanGoodfellowJitendraM.MalikVikashMansinghkaJudeaPearlMichaelWooldridgeCopyright2021,2010,2003byPearsonEducation,Inc.oritsaffiliates,221RiverStreet,Hoboken,NJ07030.AllRightsReserved.ManufacturedintheUnitedStatesofAmerica.Thispublicationisprotectedbycopyright,andpermissionshouldbeobtainedfromthepublisherpriortoanyprohibitedreproduction,storageinaretrievalsystem,ortransmissioninanyformorbyanymeans,electronic,mechanical,photocopying,recording,orotherwise.Forinformationregardingpermissions,requestforms,andtheappropriatecontactswithinthePearsonEducationGlobalRightsandPermissionsdepartment,pleaseofthird-partycontentappearontheappropriatepagewithinthetext.CoverImages:

AlanTuringScienceHistoryImages/AlamyStockPhotoStatueofAristotlePanosKaras/ShutterstockAdaLovelacePictorialPressLtd/AlamyStockPhotoAutonomouscarsAndreySuslov/ShutterstockAtlasRobotBostonDynamics,Inc.BerkeleyCampanileandGoldenGateBridgeBenChu/ShutterstockBackgroundghostednodesEugeneSergeev/AlamyStockPhotoChessboardwithchessfigureTitania/ShutterstockMarsRoverStocktrekImages,Inc./AlamyStockPhotoKasparovKATHYWILLENS/APImagesPEARSON,ALWAYSLEARNINGisanexclusivetrademarkownedbyPearsonEducation,Inc.oritsaffiliatesintheU.S.and/orothercountries.Unlessotherwiseindicatedherein,anythird-partytrademarks,logos,oriconsthatmayappearinthisworkarethepropertyoftheirrespectiveowners,andanyreferencestothird-partytrademarks,logos,icons,orothertradedressarefordemonstrativeordescriptivepurposesonly.Suchreferencesarenotintendedtoimplyanysponsorship,endorsement,authorization,orpromotionofPearsonsproductsbytheownersofsuchmarks,oranyrelationshipbetweentheownerandPearsonEducation,Inc.,oritsaffiliates,authors,licensees,ordistributors.LibraryofCongressCataloging-in-PublicationDataNames:

Russell,StuartJ.（StuartJonathan）,author.|Norvig,Peter,author.Title:

Artificialintelligence:

amodernapproach/StuartJ.RussellandPeterNorvig.Description:

Fourthedition.|Hoboken:

Pearson,2021|Series:

Pearsonseriesinartificialintelligence|Includesbibliographicalreferencesandindex.|Summary:

“UpdatededitionofpopulartextbookonArtificialIntelligence.”Providedbypublisher.Identifiers:

LCCN2019047498|ISBN9780134610993（hardcover）Subjects:

LCSH:

Artificialintelligence.Classification:

LCCQ335.R862021|DDC006.3dc23LCrecordavailableathttps:

/lccn.loc.gov/2019047498ScoutAutomatedPrintCodeISBN-10:

0-13-461099-7ISBN-13:

978-0-13-461099-3ForLoy,Gordon,Lucy,George,andIsaacS.J.R.ForKris,Isabella,andJulietP.N.PrefaceArtificialIntelligence（AI）isabigfield,andthisisabigbook.Wehavetriedtoexplorethefullbreadthofthefield,whichencompasseslogic,probability,andcontinuousmathematics;perception,reasoning,learning,andaction;fairness,trust,socialgood,andsafety;andapplicationsthatrangefrommicroelectronicdevicestoroboticplanetaryexplorerstoonlineserviceswithbillionsofusers.Thesubtitleofthisbookis“AModernApproach.”Thatmeanswehavechosentotellthestoryfromacurrentperspective.Wesynthesizewhatisnowknownintoacommonframework,recastingearlyworkusingtheideasandterminologythatareprevalenttoday.Weapologizetothosewhosesubfieldsare,asaresult,lessrecognizable.NewtothiseditionThiseditionreflectsthechangesinAIsincethelasteditionin2010:

Wefocusmoreonmachinelearningratherthanhand-craftedknowledgeengineering,duetotheincreasedavailabilityofdata,computingresources,andnewalgorithms.Deeplearning,probabilisticprogramming,andmultiagentsystemsreceiveexpandedcoverage,eachwiththeirownchapter.Thecoverageofnaturallanguageunderstanding,robotics,andcomputervisionhasbeenrevisedtoreflecttheimpactofdeeplearning.Theroboticschapternowincludesrobotsthatinteractwithhumansandtheapplicationofreinforcementlearningtorobotics.PreviouslywedefinedthegoalofAIascreatingsystemsthattrytomaximizeexpectedutility,wherethespecificutilityinformationtheobjectiveissuppliedbythehumandesignersofthesystem.NowwenolongerassumethattheobjectiveisfixedandknownbytheAIsystem;instead,thesystemmaybeuncertainaboutthetrueobjectivesofthehumansonwhosebehalfitoperates.Itmustlearnwhattomaximizeandmustfunctionappropriatelyevenwhileuncertainabouttheobjective.WeincreasecoverageoftheimpactofAIonsociety,includingthevitalissuesofethics,fairness,trust,andsafety.Wehavemovedtheexercisesfromtheendofeachchaptertoanonlinesite.Thisallowsustocontinuouslyaddto,update,andimprovetheexercises,tomeettheneedsofinstructorsandtoreflectadvancesinthefieldandinAI-relatedsoftwaretools.Overall,about25%ofthematerialinthebookisbrandnew.Theremaining75%hasbeenlargelyrewrittentopresentamoreunifiedpictureofthefield.22%ofthecitationsinthiseditionaretoworkspublishedafter2010.OverviewofthebookThemainunifyingthemeistheideaofanintelligentagent.WedefineAIasthestudyofagentsthatreceiveperceptsfromtheenvironmentandperformactions.Eachsuchagentimplementsafunctionthatmapsperceptsequencestoactions,andwecoverdifferentwaystorepresentthesefunctions,suchasreactiveagents,real-timeplanners,decision-theoreticsystems,anddeeplearningsystems.Weemphasizelearningbothasaconstructionmethodforcompetentsystemsandasawayofextendingthereachofthedesignerintounknownenvironments.Wetreatroboticsandvisionnotasindependentlydefinedproblems,butasoccurringintheserviceofachievinggoals.Westresstheimportanceofthetaskenvironmentindeterminingtheappropriateagentdesign.OurprimaryaimistoconveytheideasthathaveemergedoverthepastseventyyearsofAIresearchandthepasttwomillenniaofrelatedwork.Wehavetriedtoavoidexcessiveformalityinthepresentationoftheseideas,whileretainingprecision.Wehaveincludedmathematicalformulasandpseudocodealgorithmstomakethekeyideasconcrete;mathematicalconceptsandnotationaredescribedinAppendixAandourpseudocodeisdescribedinAppendixB.Thisbookisprimarilyintendedforuseinanundergraduatecourseorcoursesequence.Thebookhas28chapters,eachrequiringaboutaweeksworthoflectures,soworkingthroughthewholebookrequiresatwo-semestersequence.Aone-semestercoursecanuseselectedchapterstosuittheinterestsoftheinstructorandstudents.Thebookcanalsobeusedinagraduate-levelcourse（perhapswiththeadditionofsomeoftheprimarysourcessuggestedinthebibliographicalnotes）,orforself-studyorasareference.Throughoutthebook,importantpointsaremarkedwithatriangleiconinthemargin.Whereveranewtermisdefined,itisalsonotedinthemargin.Subsequentsignificantusesofthetermareinbold,butnotinthemargin.Wehaveincludedacomprehensiveindexandanextensivebibliography.TermTheonlyprerequisiteisfamiliaritywithbasicconceptsofcomputerscience（algorithms,datastructures,complexity）atasophomorelevel.Freshmancalculusandlinearalgebraareusefulforsomeofthetopics.OnlineresourcesOnlineresourcesareavailablethroughoratthebooksWebsite,aima.cs.berkeley.edu.Thereyouwillfind:

Exercises,programmingprojects,andresearchprojects.Thesearenolongerattheendofeachchapter;theyareonlineonly.Withinthebook,werefertoanonlineexercisewithanamelike“Exercise6.NARY.”InstructionsontheWebsiteallowyoutofindexercisesbynameorbytopic.ImplementationsofthealgorithmsinthebookinPython,Java,andotherprogramminglanguages（currentlyhostedatlistofover1400schoolsthathaveusedthebook,manywithlinkstoonlinecoursematerialsandsyllabi.Supplementarymaterialandlinksforstudentsandinstructors.Instructionsonhowtoreporterrorsinthebook,inthelikelyeventthatsomeexist.BookcoverThecoverdepictsthefinalpositionfromthedecisivegame6ofthe1997chessmatchinwhichtheprogramDeepBluedefeatedGarryKasparov（playingBlack）,makingthisthefirsttimeacomputerhadbeatenaworldchampioninachessmatch.Kasparovisshownatthetop.TohisrightisapivotalpositionfromthesecondgameofthehistoricGomatchbetweenformerworldchampionLeeSedolandDeepMindsALPHAGOprogram.Move37byALPHAGOviolatedcenturiesofGoorthodoxyandwasimmediatelyseenbyhumanexpertsasanembarrassingmistake,butitturnedouttobeawinningmove.AttopleftisanAtlashumanoidrobotbuiltbyBostonDynamics.Adepictionofaself-drivingcarsensingitsenvironmentappearsbetweenAdaLovelace,theworldsfirstcomputerprogrammer,andAlanTuring,whosefundamentalworkdefinedartificialintelligence.AtthebottomofthechessboardareaMarsExplorationRoverrobotandastatueofAristotle,whopioneeredthestudyoflogic;hisplanningalgorithmfromDeMotuAnimaliumappearsbehindtheauthorsnames.BehindthechessboardisaprobabilisticprogrammingmodelusedbytheUNComprehensiveNuclear-Test-BanTreatyOrganizationfordetectingnuclearexplosionsfromseismicsignals.AcknowledgmentsIttakesaglobalvillagetomakeabook.Over600peoplereadpartsofthebookandmadesuggestionsforimprovement.Thecompletelistisataima.cs.berkeley.edu/ack.html;wearegratefultoallofthem.Wehavespaceheretomentiononlyafewespeciallyimportantcontributors.Firstthecontributingwriters:

JudeaPearl（Section13.5,CausalNetworks）;VikashMansinghka（Section15.3,ProgramsasProbabilityModels）;MichaelWooldridge（Chap