PRENTICE HALL SERIES IN ARTIFICIAL INTELLIGENCE

vinegarclothΤεχνίτη Νοημοσύνη και Ρομποτική

17 Ιουλ 2012 (πριν από 5 χρόνια και 3 μήνες)

264 εμφανίσεις

PRENTICE HALL SERIES
IN ARTIFICIAL INTELLIGENCE
StuartRussellandPeterNorvig,Editors
_I
FORSYTH
&
PONCE
GRAHAM
JURAFSKY
&
MARTIN
NEAPOLITAN
RUSSELL
&
NORVIG
ComputerVision:A ModernApproach
ANSI CommonLisp
SpeechandLanguageProcessing
LearningBayesianNetworks
Artificial Intelligence:A ModernApproach
.....J.'.
.I;
ArtificialIntelligence
A ModernApproach
SecondEdition
Stuart
J.
RussellandPeterNorvig
Contributingwriters:
JohnF.Canny
DouglasD.Edwards
JitendraM.Malik
SebastianThrun
 PearsonEducationlnternational
u purchasedthisbookwithintheUnitedStatesor Canadayoushouldbeawarethatir hasbeenwrongfullyimported
:mttheapprovalof thepublisherortheAuthor.
~PresidentandEditorialDirector,ECS:MarciaJ.Horton
Ilisher:AlanR.Apt
ociateEditor:ToniDianneHolm
torialAssistant:PatrickLindner
ePresidentandDirectorof PraductionandManufacturing,ESM:DavidW.Riccardi
:cutiveManagingEditor:Vinceü'Brien
;istantManagingEditor:CamilleTrentacoste
,ductionEditor:IrwinZucker
nufacturingManager:TrudyPisciotti
nufacturingBuyer:LisaMcDowell
·ector,CreativeServices:PaulBelfanti
:ativeDirector:CaroleAnson
·Editor:GregDulles
:Director:HeatherScott
sistanttoArtDirector:GeoffreyCassar
verDesigners:StuartRussellandPeterNorvig
verlrnageCreation:StuartRussellandPeterNorvig;TamaraNewnamandPatriceVanAcker
eriorDesigner:StuartRussellandPeterNorvig
rrketingManager:PamelaShaffer
rrketingAssistant:BarrieReinhold
For Loy,Gordon,andLucy -
S.J.R.
For Kris,Isabella,andJuliet-
P.N.
11
©
2003.]995byPearsonEducarian.lnc.
PearsonEducation,lnc..
UpperSaddleRiver,NewJersey07458
II
rightsreserved.Nopartof thisbookrnaybereproduced,inanyformor byanymeans,
ithoutpermissioninwritingframthepublisher.
le authorandpublisherof thisbookhaveusedtheirbesteffortsinpreparingthisbook.Theseefforts
c1udethedevelopment,research,andtestingof thetheoriesandprograrn;todeterminetheireffectiveness.
~eauthorandpublishermakenowarrantyof anykind,expressorimplied,withregardtotbeseprograms
· thedocumentationcontainedinthisbook.Theauthorandpublishershallnotbeliableinanyeventfor
cidentalarconsequentialdamagesinconnectionwith,or arisingoutof,thefurnishing,performance,
· useof theseprograms.
rintedintheUnitedStatesof America
09876543
:SBN 0-13-080302-2
learsonEducationLtd.,London
'earsonEducarionAustraliaPty.Lld.Sydney
learsonEducationSingapore,Pte.Ltd.
'earsonEducationNorth Asia Lld.,Hang Kang
)earsonEducationCanada,Inc.,Toronto
)earsonEducacióndeMexico,S.A.de
c.v.
)earsonEducation-Japan,Tokyo
)earsonEducationMalaysia,Pte.Ltd.
)earsonEducation,lnc.,UpperSaddleRiver,NewJersey
';',
"
'..
:;."
Preface
Artificial Intelligence(AI) is a big field,andthisis a big book.We havetriedto explorethefull
breadthof thefield,whichencompasseslogic,probability,andcontinuousmathematics;perception,
reasoning,learning,andaction;andeverythingfrommicroelectronicdevicesto roboticplanetary
explorers.The bookis alsobig becausewe go into somedepthin presentingresults,althoughwe
striveto coveronly themostcentralideasin themainpartof eachchapter.Pointersaregivento
furtherresultsin thebibliographicalnotesattheendof eachchapter.
The subtitleof thisbookis"A ModernApproach."Theintendedmeaningof thisratherempty
phraseis thatwehavetriedto synthesizewhatis nowknownintoa commonframework,ratherthan
tryingto explaineachsubfieldof AI in its ownhistoricalcontext.We apologizeto thosewhose
subfieldsare,asaresult,lessrecognizablethantheymightotherwisehavebeen.
The mainunifyingthemeis theideaof anintelligentagent.We defineAI asthestudyof
agentsthatreceiveperceptsfromtheenvironmentandperformactions.Eachsuchagentimplementsa
functionthatmapsperceptsequencestoactions,andwecoverdifferentwaystorepresentthesefunc-.
tions,suchasproductionsystems,reactiveagents,real-timeconditionalplanners,neuralnetworks,
anddecision-theoreticsystems.Weexplaintheroleof learningasextendingthereachof thedesigner
iuto unknownenvironments,andwe showhowthatrole constrainsagentdesign,favoringexplicit
knowledgerepresentationandreasoning.We treatroboticsandvisionnot asindependentlydefined
problems,butasoccurringin theserviceof achievinggoals.Westresstheimportanceof thetask
environmentindeterminingtheappropriateagentdesigno
Our primaryaimis toconveytheideasthathaveemergedoverthepastfiftyyearsof AI research
andthepasttwomilleniaof relatedwork.Wehavetriedto avoidexcessiveformalityin thepresen­
tationof theseideaswhileretainingprecision.Whereverappropriate,wehaveincludedpseudocode
algorithmstomaketheideasconcrete;ourpseudocodeis describedbrieflyinAppendixB.Implemen­
tationsin severalprogramminglanguagesareavailableonthebook'sWebsite,aima.cs.berkeley.edu.
This bookis primarilyintendedfor usein anundergraduatecourseor coursesequence.It can
alsobeusedin a graduate-Ievelcourse(perhapswiththeadditionof someof theprimarysources
suggestedin thebibliographicalnotes).Becauseof its comprehensivecoverageandlargenumberof
detailedalgorithms,it is usefulasaprimaryreferencevolumefor AI graduatestudentsandprofes­
sionalswishingto branchout beyondtheirownsubfield.The only prerequisiteis familiaritywith
basicconceptsof computerscience(algorithms,datastructures,complexity)at a sophomoreleveI.
Freshmanca1culusis usefulforunderstandingneuralnetworksandstatisticallearningindetail.Some
of therequiredmathematicalbackgroundis suppliedin AppendixA.
Overviewofthebook
Thebookis dividedintoeightparts.PartI,Artificial Intelligence,offersa viewof theAI enterprise
basedaroundtheideaof intelligentagents-systemsthatcandecidewhattodoandthendoit.Part
lI,ProblemSolying,concentratesonmethodsfor decidingwhattodowhenoneneedstothinkabead
severalsteps-for examplein navigatingacrossacountryor playingchess.PartIII,Knowledgeand
Reasoning,discusseswaystorepresentknowledgeabouttheworld-how it works,whatit iscurrently
like,andwhatone'sactionsmightdo-and howto reasonlogicallywiththatknowledge.Part IV,
Planning,thendiscusseshowto usethesereasoningmethodstodecidewhatto do,particularlyby
constructingplans.PartV,Uncertain KnowledgeandReasoning,is analogousto PartsIII andIV,
butit concentratesonreasoninganddecisionmakingin thepresenceof uncertaintyabouttheworld,
asmightbefaced,for example,bya systemfor medicaldiagnosisandtreatment.
Together,PartslI-V describethatpartof theintelligentagentresponsibleforreachingdecisions.
PartVI,Learning,describesmethodsforgeneratingtheknowledgerequiredbythesedecision-making
vii
viii
Preface
components.PartVII,Communicating,Perceiving,andActing,describeswaysin whichanintel­
ligentagemcanperceiveitsenvironmentsoastoknowwhatis goingon,whetherbyvision,touch,
hearing,or understandinglanguage,andwaysinwhichit canturnitsplansimorealactions,eitheras
robotmotionor asnaturallanguageutterances.Finally,PartVIII,Conclusions,analyzesthepastand
futureof AI andthephilosophicalandethicalimplicationsof artificialintelligence.
Changesfromthefirstedition
MuchhaschangedinAI sincethepublicationof thefirsteditionin 1995,andmuchhaschangedinthis
book.Everychapterhasbeensignifieantlyrewrittentorefleetthelatestworkin thefield,toreinterpret
oldworkin a waythatis moreeohesivewithnewfindings,andtoimprovethepedagogicalflowof
ideas.Followersof AI shouldbeencouragedthateurrentteehniquesaremuchmorepraeticalthan
thoseof 1995;for exampletheplanningalgorithmsin thefirsteditioncouldgenerateplansof only
dozensof steps,whilethealgorithmsin thiseditionsealeuptotensof thousandsof steps.Similar
orders-of-magnitudeimprovementsareseeninprobabilistieinference,languageproeessing,andother
subfields.Thefollowingarethemostnotableehangesinthebook:
o InPartI,weacknowledgethehistoricaleontributionsof controltheory,gametheory,economics,
andneuroscience.This helpssetthetonefor a moreintegratedcoverageof theseideasin
subsequentehapters.
o
In PartlI,onlinesearehalgorithmsarecoveredandanewchapteronconstraintsatisfactionhas
beenadded.Thelatterprovidesanaturalconneetiontothematerialonlogic.
o In PartIII,propositionallogic,whichwaspresentedasa stepping-stoneto first-orderlogicin
thefirstedition,is nowpresemedasausefulrepresentationlanguageinitsownright,withfast
inferencealgorithmsandcircuit-basedagentdesigns.Thechapterson first-orderlogiehave
beenreorganizedtopresentthematerialmoreclearlyandwehaveaddedtheInternetshopping
domainasanexample.
o
In PartIV,weincludenewerplanningmethodssuchasGRA~HPLANandsatisfiability-based
planning,andweincreaseeoverageof scheduling,conditionalplanning,hierarehiealplanning,
andmultiagentplanning.
o In PartV,wehaveaugmentedth~materialon Bayes,iannetworkswithnewalgorithms,sueh
asvariableeliminationandMarkovChainMonteCarlo,andwehaveereateda newchapteron
uncertaintemporalreasoning,coveringhiddenMarkovmodels,Kalmanfilters,anddynamic
Bayesiannetworks.Theeoverageof Markovdecisionprocessesis deepened,andweaddsec­
tionsongametheoryandmeehanismdesigno
o In PartVI,wetietogetherworkin statistical,symbolie,andneurallearningandaddsectionson
boostingalgorithms,theEM algorithm,instance-basedlearning,andkernelmethods(support
vectormaehines).
o In PartVII,eoverageof languageprocessingaddssectionsondiseourseprocessingandgram­
marinduetion,aswell asa ehapteronprobabilisticlanguagemodels,withapplicationstoin­
formationretrievalandmaehinetranslation.Thecoy,erage\lf robotiesstressestheintegrationof
uncertainsensordata,andtheehapteronvisionhasúpdatedmaterialonobjeetrecognition.
o In PartVIII,weintroduceasectionontheethicalimplicatioris"OfAI.
Usingthisbook
Thebookhas27chapters,eachrequiringabouta week'sworthof leetures,soworkingthroughthe
wholebookrequiresatwo-semestersequence.Alternatively,aeourseeanbetailoredtosuittheinter­
estsof theinstructorandstudent.Throughitsbroadcoverage,thebookcanbeusedtosupportsuch
Preface
1~!~
~
~
NEW TERM
ix
courses,whethertheyareshort,introductoryundergraduateeoursesor specializedgraduatecourseson
advaneedtopies.Samplesyllabifromthemorethan600universitiesandeollegesthathaveadopted
thefirsteditionareshownontheWebataima.cs.berkeley.edu,alongwithsuggestionstohelpyoufind
asequeneeappropriatetoyourneeds.
Thebookincludes385exereises.Exercisesrequiringsignificantprogrammingaremarkedwith
a keyboardicon.Theseexercisescanbestbesolvedby takingadvantageof thecoderepositoryat
aima.cs.berkeley.edu.Someof themarelargeenoughtobeconsideredtermprojects.A numberof
exercisesrequiresomeinvestigationof theliterature;thesearemarkedwithabookicon.
Throughoutthebook,importantpointsaremarkedwithapointingicon.Wehaveincludedan
extensiveindexof around10,000itemstomakeit easytofindthingsin thebook.Wherevera new'
termisfirstdefined,it isalsomarkedin themargin.
UsingtheWeb
site
At theaima.cs.berkeley.eduWebsiteyouwill find:
o
implementatiónsof thealgorithmsinthebookinseveralprogramminglanguages,
o alist of over600schoolsthathaveusedthebook,manywithlinkstoonlineeoursemateriais,
o anannotatedlistof over800linkstositesaroundthewebwithusefulAI content,
o aehapterbychapter'listof supplementarymaterialandlinks,
o instructionsonhowtojoin adiseussiongroupforthebook,
o instructionsonhowtocontaettheauthorswithquestionsor cornrnents,
o
instructionsonhowtoreporterrorsinthebook,in thelikelyeventthatsomeexist,and
o eopiesof thefiguresinthebook,alongwithslidesandothermaterialforinstructors.
Acknowledgments
JitendraMalik wrotemostof Chapter24(onvision).Most of Chapter25(onroboties)waswritten
by SebastianThrunin thiseditionandbyJohnCannyin thefirstedition.DougEdwardsresearehed
thehistoriealnotesfor thefirstedition.TimHuang,MarkPaskin,andCynthiaBruynshelpedwith
formattingof thediagramsandalgorithms.AlanApt,SondraChavez,Toni Holm,JakeWarde,Irwin
Zucker,andCamilleTrentacosteat PrenticeHall triedtheirbesttokeepus on scheduleandmade
manyhelpfulsuggestionsonthebook'sdesignandcontent.
Stuartwouldliketothankhis parentsfor theircontinuedsupportandencouragementandhis
wife,Loy Sheflott,forherendlesspatienceandboundlesswisdom.HehopesthatGordonandLucy
will soonbereadingthis.RUGS (Russell'sUnusualGroupof Students)havebeenunusuallyhelpful.
Peterwouldlike tothankhisparents(TorstenandGerda)forgettinghimstarted,andhiswife
(Kris),ehildren,andfriendsfor eneouragingandtoleratinghimthroughthelonghoursof writingand
longerhoursof rewriting.
WeareitidebtedtothelibrariansatBerkeley,Stanford,MIT,andNASA,andtothedevelopers
of CiteSeerandGoogle,whohaverevolutionizedthewaywedoreseareh.
We can'tthankali thepeoplewhohaveusedthebookandmadesuggestions,butwe would
liketoacknowledgetheespeciallyheJpfulcornrnentsof Eyal Amit,KrzysztofApt,ElleryAziel,Jeff
VanBaalen,BrianBaker,DonBarker,TonyBarrett,JamesNewtonBass,DonBeal,HowardBeck,
WolfgangBibel,JohnBinder,LarryBookman,DavidR.Boxall,GerhardBrewka,SelmerBringsjord,
CarlaBrodley,ChrisBrown,WilhelmBurger,LaurenBurka,JoaoCachopo,MurrayCampbell,Nor­
manCarver,EmmanuelCastro,Anil Chakravarthy,DanChisarick,RobertoCipolla,DavidCohen,
JamesColeman,Julie Ann Comparini,GaryCottrell,ErnestDavis,RinaDeehter,TomDietterich,
ChuckDyer,BarbaraEngelhardt,DougEdwards,KutluhanErol,OrenEtzioni,HanaFilip,Douglas
x
Preface
Fisher,JeffreyForbes,KenFord,JohnFosler,Alex Franz,BobFutrelle,MarekGalecki,StefanGer­
berding,StuartGill,SabineGlesner,SethGolub,GostaGrahne,RussGreiner,Eric Grimson,Barbara
Grosz,LarryHall,SteveHanks,OtharHansson,ErnstHeinz,JimHendler,ChristophHerrmann,Vas­
antHonavar,Tim Huang,SethHutchinson,Joost Jacob,!vlagnusJohansson,DanJurafsky,Leslie
Kaelbling,Keiji Kanazawa,SurekhaKasibhatla,SimonKasif,HenryKautz,GernotKerschbaumer,
RichardKirby,KevinKnight,SvenKoenig,DaphneKoller,RichKorf,J amesKurien,J ohnLafferty,
Gus Larsson,John Lazzaro,Jon LeBlanc,JasonLeatherman,FrankLee,EdwardLim,PierreLou­
veaux,DonLoveland,SridharMahadevan,JimMartin,Andy!vlayer,David!vlcGrane,Jay!vlendel­
sohn,Brian!vli1ch,SteveMinton,VibhuMittal,LeoraMorgenstem,StephenMuggleton,KevinMur­
phy,RonMusick,SungMyaeng,LeeNaish,PanduNayak,BernhardNebel,StuartNelson,XuanLong
Nguyen,Illah Nourbakhsh,SteveOmohundro,DavidPage,DavidPalmer,DavidParkes,RonParr,
MarkPaskin,TonyPassera,MichaelPazzani,WimPijls,IraPohl,MarthaPollack,DavidPoole,Bruce
Porter,Ma1colmPradhan,Bill Pringle,LorrainePrior,GregProvan,WilliamRapaport,PhilipResnik,
FrancescaRossi,J onathanSchaeffer,RichardScherl,Lars Schuster,SoheilShams,StuartShapiro,
Jude Shavlik,SatinderSingh,Daniel Sleator,DavidSmith,BryanSo,RobertSproull,LynnStein,
.,Larry Stephens,AndreasSto1cke,Paul Stradling,DevikaSubramanian,Rich Sutton,JonathanTash,
,"AustinTate,!vlichaelThielscher,WilliamThompson,SebastianThrun,Eric Tiedemann,!vlarkTor­
rance,RandallUpham,Paul Utgoff,PetervanBeek,Hal Varian,Sunil Vemuri,JimWaldo,Bonnie
Webber,DanWeld,!vlichaelWellman,MichaelDeanWhite,KaminWhitehouse,BrianWilliams,
DavidWolfe,Bill Woods,AldenWright,RichardYen,WeixiongZhang,ShlomoZilberstein,andthe
anonymousreviewersprovidedbyPrenticeHall.
AbouttheCover
Thecoverimagewasdesignedby theauthorsandexecutedby Lisa MarieSardegnaandMaryann
SimmonsusingSGI Inventor andAdobePhotoshopTrvIThecoverdepictsthefollowingitems
fromthehistoryof AI:
1.Aristotle'splanningalgorithmfromDeMotllAnimalillm(c.400
B.C.).
2.RamonLull's conceptgeneratorfromArsMagna(c.1300
A.D.).
3.CharlesBabbage'sDifferenceEngíne,aprototypefor thefirstuniversalcomputer(1848).
4.GottlobFrege'snotationforfirst-orderlogic(1789).
5.LewisCarroll'sdiagramsfor logicalreasoning(1886).
6.SewallWright'sprobabilisticnetworknotation(1921).
7.AlanTuring(1912-1954).
8.ShakeytheRobot(1969-1973).
9.A moderndiagnôsticexpertsystem(1993).
;.
AbouttheAuthors
StuartRussellwasbomin 1962in Portsmouth,England.He receivedhisB.A.withfirst-classhon­
oursin physicsfromOxfordUniversityin 1982,andbisPh.D.in computersciencefromStanfordin
1986.He thenjoinedthefacultyof theUniversityof CaliforniaatBerkeley,whereheis aprofessor
of computerscience,directorof theCenterfor IntelligentSystems,andholderof theSmith-Zadeh
Chairin Engineering.In 1990,hereceivedthePresidentialYoungInvestigatorAwardof theNational
ScienceFoundation,andin 1995hewascowinnerof theComputersandThoughtAward.He wasa
1996Miller Professorof theUniversityof CalifomiaandwasappointedtoaChancellor'sProfessor­
shipin2000.In 1998,hegavetheForsytheMemoria1LecturesatStanfordUniversity.He is aFellow
andformerExecutiveCouncilmemberof theAmericanAssociationforArtificialIntelligence.Hehas
publishedover100papersonawiderangeof topicsinartificialintelligence.His otherbooksinclude
TheUseof KnowledgeinAnalogyandIndllctionand(withEric Wefald)Do theRightThing:Stlldies
inLimitedRationality.
PeterNorvigis directorof SearchQualityat Google,Inc.He is a FellowandEXifutive Council
memberof theAmericanAssociationforArtificialIntelligence.Previous1y,hewasheadof theCom­
putationalSciencesDivisionatNASA AmesResearchCenter,whereheoversawNASA's research
anddevelopmentinartificialintelligenceandrobotics.BeforethatheservedaschiefscientistatJun­
glee,wherehehelpeddeveloponeof thefirstInternetinformationextractionservices,andasasenior
scientistatSunMicrosystemsLaboratoriesworkingonintelligentinformationretrieval.He received
aB.S.in appliedmathematicsfromBrownUniversityandaPh.D.in computersciencefromtheUni­
versityof CalifomiaatBerkeley.HehasbeenaprofessorattheUniversityof SouthemCalifomiaand
a researchfacultymemberatBerkeley.He hasover50publicationsin computerscienceincluding
thebooksParadigmsof AI Programming:CaseStlldiesin CommonLisp,Verbmobil:A Translation
SystemforFace-to-FaceDialog,andIntelligentHelpSystemsfor UNIX.
xi
"i-<
Summaryaf Cantents
I ArtificialInteIligence
1 Introduction................................................................. 1
2 IntelligentAgents 32
II Problem-solving
3 SolvingProblemsbySearching.............................................59
4 InformedSearchandExploration 94
5 ConstraintSatisfactionProblems..............................................137
6 Adversarial.Search........................................................... 161
III Knowledgeandreasoning
7 Logical Agents 194
8 First-Order Logic............................................................. 240
9 Inferencein First-Order Logic 272
10 KnowledgeRepresentation 320
IV Planning
11 Planning 375
12 PlanningandActingin theReal World.....................................417
V Uncertainknowledgeandreasoning
13 Uncertainty.............................................................462
14 ProbabilisticReasoning....................................................492
15 ProbabilisticReasoningoverTime..........................................537
16 Making SimpleDecisions..................................................584
17 Making ComplexDecisions...............................................613
VI Learning
18 LearningfromObservations..............................................649
19 Knowledgein Learning...........................................678
20 StatisticalLearningMethods........................................712
21 ReinforcementLearning.............................................763
VII Communicating,perceiving,andacting
22 Communication........................................................
ó 
790·
23 ProbabilisticLanguageProcessing.............................................834
24 Perception..............................................................863
25 Robotics 901
VIII Conclusions
26 PhilosophicalFoundations................................................947
27 AI:PresentandFuture......................................................968
A lVlathematicalbackground 977
B NotesonLanguagesandAIgorithms....................................... 984
Bibliography 987
Index· 1045
xiii
Contents
I ArtificialIntelligence
,.
./
,~
1 Introduction
l.l What is AI?.
Acting humanly:The Turing Test approach.
Thinking humanly:The cognitive modeling approach
Thinking rationally:The"Iaws of thought"approach.
Acting rationally:The rational agent approach
1.2 The Foundations of Artificial Intelligence.
Philosophy (428 B.C.-present)
Mathematics (c.800-present)
Economics (l776-present).
Neuroscience (l861-present)
Psychology (1879-present).
Computer engineering (l940-present)
Control theory andCybernetics (l948-present).
Linguistics (l957-present).
1.3 The History of Artificial Intelligence.
The gestationof artificial intelligence (1943-1955).
The birth of artificial intelligence (1956).....
Early enthusiasm,great expectations (1952-1969).
A doseof reality (1966-1973).
Knowledge-based systems:The key to power?(1969-1979).
AI becomes anindustry (l980-present)....
The returnof neural networks (l986-present).
AI becomes ascience (l987-present).
The emergenceof intelligent agents(l995-present)
IA
The Stateof theArt.
1.5 Summary.
Bibliographical andHistorical Notes
Exercises.
2 IntelligentAgents
2.1 Agents andEnvironments.
2.2 Good Behavior:The Concept of Rationality
Performance measures.
Rationality.
Omniscience,learning,andautonomy.
2.3 The Natureof Environments...
Specifying thetask environment.
Prop.ertiesof task environments
204
The StrUctureof Agents
Agent programs.
Simple reflex agents.
Model-based reflex agents
xv
1
I
2
3
4
4
5
5
7
9
10
12
14
15
16
16
16
17
18
21
22
24
25
25
27
27
28
29
30
32
32
34
35
35
36
38
38
40
44
44
46
48
xvi
Goal-basedagents.
Utility-basedagents
Learningagents..
2.5 Summary.
BibliographicalandHistoricalNotes
Exercises.
II Problem-solving
3 SolvingProblemsbySearching
3.1 Prob1em-SolvingAgents.
Well-definedproblemsandsolutions
Formulatingproblems
3.2'ExampleProblems..
Toyprob1ems
Real-worldproblems
3.3 SearchingforSolutions
Measuringproblem-solvingperformance
3.4 UninformedSearchStrategies
Breadth-firstsearch.
Depth-firstsearch.
Depth-limitedsearch.
lterativedeepeningdepth-firstsearch
Bidirectionalsearch.
Comparinguninformedsearchstrategies
3.5 AvoidingRepeatedStates.....
3.6 SearchingwithPartia!lnformation
Sensor1essproblems..
Contingencyproblems.
3.7 Summary.
BibliographicalandHistoricalNotes'.
Exercises.
4 InformedSearchandExp1oration
4.1 lnformed(Heuristic)SearchStrategies.
Greedybest-firstsearch.
A* search:Minimizingthetotalestimatedsolutioncost
Memory-boundedheuristicsearch
Learningtosearchbetter.
4.2 HeuristicFunctions.
Theeffectof heuristicaccuracyonperformance
lnventingadmissib1eheuristicfunctions
!,..
r
~>
Learningheuristicsfromexperience.
4.3 LocalSearchAlgorithmsandOptimizationProblems
Hill-climbingsearch....
Simulatedannealingsearch
Local beamsearch.....
Geneticalgorithms.
4.4 Local SearchinContinuousSpaces
Contents
49
51
51
54
55
56
59
59
62
62
64
64
67
69
71
73
73
75
77
78
79
8i
81
83
84
86
87
88
89
94
94
95
97
101
104
105
106
107
109
110
i11
115
115
116
119
Contents
4.5 OnlineSearchAgentsandUnknownEnvironments
Onlinesearchproblems
Onlinesearchagents..
Onlinelocalsearch
Leaminginonlinesearch
4.6 Summary.
BibliographicalandHistoricalNotes
Exercises.
5 ConstraintSatisfactionProb1ems
5.1 ConstraintSatisfactionProb1ems
5.2 BacktrackingSearchfor CSPs..
Variableandvalueordering...
Propagatinginformationthroughconstraints
lntelligentbacktracking:lookingbackward.
5.3 Local SearchforConstraintSatisfactionProb1ems
5.4 TheStructureof Prob1ems..
5.5 Summary.
Bibliographica1andHistoricalNotes
Exercises.
6 AdversarialSearch
6.1 Games.
6.2 OptimalDecisionsinGames
Optima1strategies.
Theminimaxalgorithm.
Optima!decisionsinmu!tiplayergames.
6.3 Alpha-BetaPruning.
6.4 lmperfect,Real-TimeDecisions.
Evaluationfunctions.
Cuttingoff search.
6.5 GamesThatlncludeanElementof Chance
Positionevaluationingameswithchancenodes
Comp1exityof expectiminimax
Cardgames.
6.6 State-of-the-ArtGamePrograms
6.7 Discussion.
6.8 Summary.
BibliographicalandHistorica1Notes
Exercises.'.
III
Knowledgeandreasoning
7 Logical Agents
7.1 Knowledge-BasedAgents
7.2 TheWumpusWorld.'..
7.3 Logic.
7.4 PropositionalLogic:A VerySimpleLogic
Syntax.
xvii
122
123
125
126
127
129
130
134
137
137
141
143
144
148
150
151
155
156
158
161
161
162
163
165
165
167
171
171
173
175
177
177
179
180
183
185
186
189
194
195
197
200
204
204
Semantics.
A simpleknowledgebase
Inference.
Equivalence,validity,andsatisfiability
7.5 ReasoningPatternsinPropositionalLogic
Resolution.
Forwardandbackwardchaining..
7.6 Effectivepropositionalinference.
A completebacktrackingalgorithm
Local-searchalgorithms.
Hardsatisfiabilityproblems.
7.7 AgentsBasedonPropositionalLogic
Findingpitsandwumpusesusinglogicalinference.
Keepingtrackof locationandorientation
Circuit-basedagents
A comparison.
7.8 Summary.
BibliographicalandHistoricalNotes
Exercises.....
8 First-Order Logic
8.1 RepresentationRevisited.
8.2 SyntaxandSemanticsof First-OrderLogic
Modelsforfirst-orderlogic
Symbolsandinterpretations
Terms.
Atomicsentences.
Complexsentences
Quantifiers.
Equality.
8.3 UsingFirst-OrderLogic
Assertionsandqueriesinfirst-orderlogic.
Thekinshipdomain..
Numbers,sets,andlists
Thewumpusworld
8.4 KnowledgeEngineeringinFirst-OrderLogic.
Theknowledgeengineeringprocess.
Theelectroniccircuitsdomain.
8.5 Summary.
BibliographicalandHistoricalNotes
Exercises.
xviii
9 Ioferencelo Flrst-Order Loglc
9.1 Propositionalvs.First-OrderInference
InferencemIesforquantifiers....
Reductiontopropositionalinference
9.2 UnificationandLifting
A first-orderinferencemie
Unification.
~,
.
~
Contents
206
208
208
210
211
213
217
220
221
222
224
225
225
227
227
231
232
233
236
240
240
245
245
246
248
248
249
249
253
253
253
254
256
258
260
261
262
266
267
268
272
272
273
274
275
275
276
Contents
Storageandretrieval.
9.3 ForwardChaining.
First-orderdefiniteclauses.
A simpleforward-chainingalgorithm
Efficientforwardchaining.
9.4 BackwardChaining.
A backwardchainingalgorithm
Logicprograrnrning.
Efficientimplementationof logicprograms
Redundantinferenceandinfiniteloops
Constraintlogicprogramming.
9.5 Resolution.
Conjunctivenormalformfor first-orderlogic
Theresolutioninferencemie
Exampleproofs.
Completenessof resolution
Dealingwithequality
Resolutionstrategies.
Theoremprovers.
9.6 Summary.
BibliographicalandHistoricalNotes
Exercises.
10KnowledgeRepresentation
10.1 OntologicalEngineering.
10.2 CategoriesandObjects
Physicalcomposition.
Measurements.....
Substancesandobjects
10.3 Actions,Situations,andEvents
Theontologyof situationcalculus.
Describingactionsin situationcalculus
Solvingtherepresentationalframeproblem.
Solvingtheinferentialframeproblem.
Timeandeventcalculus
Generalizedevents.
Processes.
Intervals.
Fluentsandobjects
10.4 MentalEventsandMentalObjects
A formaltheoryof beliefs..
Knowledgeandbelief....
Knowledge,time,andaction
10.5 TheInternetShoppingWorld
Comparingoffers.
10.6 ReasoningSystemsfor Categories
Semanticnetworks.
Descriptionlogics.
10.7 ReasoningwithDefaultInformation
xix
278
280
280
281
283
287
287
289
290
292
294
295
295
297
297
300
303
304
306
310
310
315
320
320
322
324
325
327
328
329
330
332
333
334
335
337
338
339
341
341
343
344
344
348
349
350
353
354
','
13.4
13.5
13.6
xx
Openandclosedworlds.
Negationasfailureandstablemodelsemantics.
Circumscriptionanddefaultlogic
10.8 TmthMaintenanceSystems.
10,9 Summary.
BibliographicalandHistoricalNotes
Exercises.
IV Planning
11 Planning
lI.!ThePlanningProblem.
Thelanguageof planningproblems
Expressivenessandextensions
Example:Air cargotransport..
Example:Thesparetireproblem
Example:Theblocksworld...
11.2 PlanningwithState-SpaceSearch
Forwardstate-spacesearch...
Backwardstate-spacesearch..
Heuristicsfor state-spacesearch.
11.3 Partial-OrderPlanning.
A partial-orderplanningexample
Partial-orderplanningwithunboundvariables
Heuristicsforpartial-orderplanning
11.4 PlanningGraphs.
Planninggraphsforheuristicestimation
TheGRAPHPLANalgorithm.
Terminationof GRAPHPLAN.
11.5 PlanningwithPropositionalLogic'"
Describingplanningproblemsin propositionallogic
Complexityof propositionalencodings
11.6 Analysisof PlanningApproaches
li.7 Summary.
BibliographicalandHistoricalNotes
Exercises.
12 PlanningandActingin theReal World
12.1 Time,Schedules,andResources..
Schedulingwithresourceconstraints
12.2 HierarchicalTaskNetworkPlanning
Representingactiondecompositions
Modifyingtheplannerfordecompositions
Discussion.
12.3 PlanningandActinginNondeterministicDomains
12.4 ConditionalPlanning.
Conditionalplanninginfully observableenvironments.
Conditionalplanninginpartiallyobservableenvironments
12.5 ExecutionMonitoringandReplanning.
Contents
354
356
358
360
362
363
369
375
375
377
378
380
381
381
382
382
384
386
387
391
393
394
395
397
398
401
402
402
405
407
408
409
412
417
417
420
422
423
425
427
430
433
433
437
441
Contents
12.6 ContinuousPlanning.
12.7 MultiAgentPlanning.
Cooperation:Joint goalsandplans
Multibodyplanning...
Coordinationmechanisms
Competition.
12.8 Summary.
BibliographicalandHistoricalNotes
Exercises.
V Uncertainknowledgeandreasoning
13 Uncertainty
13.1 ActingunderUncertainty
Handlinguncertainknowledge
Uncertaintyandrationaldecisions.
Designfor adecision-theoreticagent
13.2 BasicProbabilityNotation.
Propositions..
AtolIÚcevents.
Prior probability.
Conditionalprobability
13.3 TheAxiomsof Probability
Usingtheaxiomsof probability
Whytheaxiomsof probabilityarereasonable
InferenceUsingFul!Joint Distributions.
Independence.
Bayes'RuleandItsUse.
ApplyingBayes'mIe:Thesimplecase.
UsingBayes'mIe:Combiningevidence
13.7 TheWumpusWorldRevisited.
13.8 Summary.
BibliographicalandHistoricalNotes
Exercises.
14 ProbabilisticReasoning
14.1 RepresentingKnowledgeinanUncertainDomain
14.2 TheSemanticsof BayesianNetworks.
Representingtheful!joint distribution.
ConditionalindependencerelationsinBayesiannetworks
14.3 EfficientRepresentationof ConditionalDistributions.
14.4 ExactInferencein BayesianNetworks
Inferencebyenumeration.....
ThevariableelilIÚnationalgorithm
Thecomplexityof exactinference.
Clusteringalgorithms.
14.5 ApproximateInferenceinBayesianNetworks
Directsamplingmethods.
InferencebyMarkovchainsimulation....
xxi
445
449
450
451
452
454
454
455
459
462
462
463
465
466
466
467
468
468
470
471
473
473
475
477
479
480
481
483
486
487
489
492
492
495
495
499
500
504
504
507
509
510
511
511
516
14.6 ExtendingProbabilitytoFirst-OrderRepresentations
14.7 OtherApproachestoUncertainReasoning.
Rule-basedmethodsfor uncertainreasoning.
Representingignorance:Dempster-Shafertheory.
Representingvagueness:Fuzzysetsandfuzzylogic
14.8 Summary.
BibliographicalandHistoricalNotes
Exercises.
xxii
15 ProbabilisticReasoningoverTime
15.1 TimeandUncertainty.
Statesandobservations.
StationaryprocessesandtheMarkovassumption.
15.2 InferenceinTemporalModels.
Filteringandprediction.
Smoothing.
Findingthemostlikelysequence
15.3 HiddenMarkovModels...
Simplifiedmatrixalgorithms..
15.4 KalmanFilters.
UpdatingGaussiandistributions.
A simpleone-dimensionalexample
Thegeneralcase.
Applicabilityof Kalmanfiltering
15.5 DynamicBayesianNetworks
ConstructingDBNs.
ExactinferenceinDBNs.
ApproximateinferenceinDBNs
15.6 SpeechRecognition
Speechsounds
Words.
Sentences
Buildingaspeechrecognizer
15.7 Summary.
BibliographicalandHistoricalNotes
Exercises.
16 Making SimpleDecisions
16.1 CombiningBeliefsandDesiresunderUncertainty
16.2 TheBasisof UtilityTheory....
Constraintsonrationalpreferences
And thentherewasUtility;.
16.3 UtilityFunctions.
Theutilityof money.
Utilityscalesandutilityassessment.
16.4 MultiattributeUtilityFunctions.
Dominance.
Preferencestructureandmultiattributeutility.
16.5 DecisionNetworks.
')
0/
o.~ o
Contents
519
523
524
525
526
528
528
533
537
537
538
538
541
542
544
547
549
549
551
553
554
556
557
559
560
563
565
568
570
572
574
576
578
578
581
584
584
586
586
588
589
589
591
593
594
596
597
Contents
Representingadecisionproblemwithadecisionnetwork
Evaluatingdecisionnetworks
16.6 TheValueofInformation
A simpleexample.
A generalformula
Propertiesof thevalueof information
Implementinganinformation-gatheringagent
16.7 Decision-TheoreticExpertSystems
16.8 Summary.
BibliographicalandHistoricalNotes
Exercises.
-17 Making ComplexDecisions
17.1 SequentialDecisionProblems.
An example.
Optimalityinsequentialdecisionproblems
17.2 ValueIteration.
Utilitiesof states.
Thevalueiterationalgorithm.
Convergenceof valueiteration
17.3 PolicyIteration.
17.4 PartiallyobservableMDPs..
17.5 Decision-TheoreticAgents..
17.6 DecisionswithMultipleAgents:GameTheory.
17.7 MechanismDesign.
17.8 Summary.
BibliographicalandHistoricalNotes
Exercises.
VI
Learning
18 LearningfromObservations
18.1 FormsofLeaming.
18.2 InductiveLearning.
18.3 LearningDecisionTrees.
Decisiontreesasperformanceelements.
Expressivenessof decisiontrees....
Inducingdecisiontreesfromexamples.
Choosingattributetests.
Assessingtheperformanceof thelearningalgorithm
Noiseandoverfitting.
Broadeningtheapplicabilityof decisiontrees.
18.4 EnsembleLearning.
18.5 WhyLearningWorks:ComputationalLearningTheory
Howmanyexamplesareneeded?
Learningdecisionlists
Discussion.
18.6 Summary.
BibliographicalandHistoricalNotes
xxiii
598
599
600
600
601
602
603
604
607
607
609
613
613
613
616
618
619
620
620
624
625
629
631
640
643
644
646
649
649
651
653
653
655
655
659
660
661
663
664
668
669
670
672
673
674
l
xxiv
Exercises.
19 Knowledgein Learning
19.1 A LogicalFormulationof Leaming
Examplesandhypotheses...
Current-best-hypothesissearch
Least-commitmentsearch
19.2 KnowledgeinLearning
Somesimpleexamples
Somegeneralschemes.
19.3 Explanation-BasedLeaming
Extractinggeneralrulesframexamples
lmpravingefficiency.
19.4 LeamingUsingRelevancelnformation
Deterrniningthehypothesisspace...
Learningandusingrelevanceinformation
19.5 lnductiveLogicProgramming.
An example.
Top-downinductivelearningmethods.
lnductivelearningwithinversededuction.
Makingdiscoverieswithinductivelogicprogramming
19.6 Summary.
BibliographicalandHistoricalNotes
Exercises.
Contents
676
678
678
678
680
683
686
687
688
690
691
693
694
695
695
697
699
701
703
705
707
708
710
Contents
20.7 CaseStudy:HandwrittenDigitRecognition
20.8 Summary.
BibliographicalandHistoricalNotes
Exercises.
21 ReinforcementLearning
21.1 lntroduction.
21.2 PassiveReinforcementLeaming
Directutilityestimation.
Adaptivedynamicpragramming
Temporaldifferencelearning.
21.3 ActiveReinforcementLeaming.
Exploration.
LeaminganAction-ValueFunction
21.4 GeneralizationinReinforcementLearning
Applicationstogame-playing
ApplicationtorobotcontraI
21.5 PolicySearch.
21.6 Summary.
BibliographicalandHistoricalNotes
Exercises.
VII Communicating,perceiving,andacting
xxv
752
754
755
759
763
763
765
766
767
767
771
771
775
777
780
780
781
784
785
788
20 StatisticalLearningMethods
20.1
20.2
StatisticalLearning.
LearningwithCompleteData.'.
Maximum-likelihoodparameterleaming:Discretemodels.
NaiveBayesmodels,.
Maximum-likelihoodparameterlearning:Continuousmodels
Bayesianparameterlearning.
LearningBayesnetstructures.
LearningwithHiddenVariables:TheEM AIgorithm.
Unsupervisedclustering:Learningmixturesof Gaussians
LeamingBayesiannetworkswithhiddenvariables
LearninghiddenMarkovmodels.
Thegeneralforrnof theEM algorithm.
LearningBayesnetstructureswithhiddenvariables
lnstance-BasedLearning
Nearest-neighbormodels
Kernelmodels.
NeuralNetworks.
Unitsinneuralnetworks
Networkstructures....
Singlelayerfeed-forwardneuralnetworks(perceptrons)
Multilayerfeed~forwardneuralnetworks
Leamingneuralnetworkstructures
KernelMachines.
20.3
20.4
20.5
20.6
~,
...
~:'
._l -.;
712
712
716
716
718
719
720
722
724
725
727
731
731
732
733
733
735
736
737
738
740
744
748
749
22 Communication
22.1 CommunicationasAction.
Fundamentaisoflanguage.
Thecomponentstepsof communication
22.2 A FormalGrammarforaFragmentof English
TheLexiconof
$0,.
TheGrammarof [o.
22.3 SyntacticAnalysis(Parsing).
Efficientparsing.
22.4 AugmentedGrammars.
Verbsubcategorization.
Generativecapacityof augmentedgrammars
22.5 SemanticInterpretation.
Thesemanticsof anEnglishfragment.
Timeandtense.
Quantification.
Pragmaticlnterpretation.
LanguagegenerationwithDCGs
22.6 AmbiguityandDisambiguation
Disambiguation.....
22.7 DiscourseUnderstanding.
Referenceresolution.
Thestructureof coherentdiscourse
22.8 Grammarlnduction
22.9 Summary.
790
790
791
792
795
795
796
798
800
806
808
809
810
811
812
813
815
817
818
820
821
821
823
824
826
~
..
'
xxvi
BibliographicalandHistoricalNotes
Exercises.
23 ProbabilisticLanguageProcessing
23.1 ProbabilisticLanguageModels
Probabilisticcontext-freegrammars.
LearningprobabilitiesforPCFGs.
LearningmIestructureforPCFGs
23.2 InformationRetrieval
EvaluatingIR systems..
IR refinements....
Presentationof resultsets
ImplementingIR systems
23.3 InformationExtraction..
23.4 MachineTranslation...
Machinetranslationsystems.
Statisticalmachinetranslation
Learningprobabilitiesformachinetranslation
23.5 Summary.
BibliographicalandHistoricalNotes
Exercises.
24 Perception
24.1 Introduction.
24.2 ImageFormation.
Imageswithoutlenses:thepinholecamera
Lenssystems.
Light:thephotometryof imageformation
Color:thespectrophotometryof imageformation
24.3 EarlyImageProcessingOperations
Edgedetection'.
Imagesegmentation.
24.4 ExtractingThree-DimensionalInformation
Motion.
Binocularstereopsis
Texturegradients.
Shading.
Contour:.
24.5 ObjectRecognition
Brightness-basedrecognition
Feature-basedrecognition..
PoseEstimation.
24.6 UsingVisionforManipulationandNavigation
24.7 Summary....
BibliographicalandHistüricalNotes
Exercises.
25 Robotics
25.1 Introduction
Contents
I
Contents
-
827
831
834
834836
839
840840
842844
845846
848
850
852
853
856
857858
861
863
863865
865866
867
868
869
870872
873875
876
879
880881
885
887
.,~~!
888
.....
,
..
,
.......
890
~~
.............,
892
894
895898
901
901
25.2 RobotHardware
Sensors.
Effectors.
25.3 RoboticPerception.
Localization.
Mapping.
Othertypesof perception
25.4 PlanningtoMove.
Configurationspace.
Cell decompositionmethods.
Skeletonizationmethods...
25.5 Planninguncertainmovements
Robustmethods.
25.6 Moving.
Dynamicsandcontrol
Potentialfieldcontrol
Reactivecontrol...
25.7 RoboticSoftwareArchitectures
Subsumptionarchitecture.
Three-Iayerarchitecture.
Roboticprogramminglanguages
25.8 ApplicationDomains.
25.9 Summary.
BibliographicalandHistoricalNotes
Exercises.
VIII ConcIusions
26 PhilosophicalFoundations
26.1 WeakAI:CanMachinesAct Intelligently?
Theargumentfromdisability.
Tbemathematicalobjection.
Tbeargumentfrominformality.
26.2 StrongAI:CanMachinesReallyThink?
Themind-bodyproblem.....
The"braininavaI"experiment.
Thebrainprosthesisexperiment.
TheChineseroom.
26.3 TheEthicsandRisksof DevelopingArtificialIntelligence
26.4 Summary.
BibliographicalandHistoricalNotes
Exercises.
27 AI:PresentandFuture
27.1 AgentComponents.
27.2 AgentArchitectures.
27.3 Are WeGoingintheRightDirection?
xxvii
903
903
904
907
908
913
915
916
916
919
922
923
924
926
927
929
930
932
932
933
934
935
938
939
942
947
947
948
949
950
952
954
955
956
958
960
964
964
967
968
968
970
972
xxviii Contents
27.4 Whatif AI DoesSucceed?974
A Mathematicalbackground 977
A.l ComplexityAna1ysisand00 Notation 977
Asymptoticanalysis 977
NP andinherentlyhardproblems 978
A.2 Vectors,Matrices,andLinearAlgebra 979
A.3 ProbabilityDistributions 981
BibliographicalandHistoricalNotes 983
B NotesonLanguagesandAlgorithms 984
B.l DefiningLanguageswithBackus-NaurForm(BNF).984
B.2 DescribingAlgorithmswithPseudocode 985
B.3 OnlineHelp 985
Bibliography 987
Index 1045
'''y.,'.