计算机科学概论原版课件(第九版)-10

Chapter10ArtificialIntelligence©2007PearsonAddison-Wesley.Allrightsreserved©2007PearsonAddison-Wesley.Allrightsreserved0-2Chapter10:ArtificialIntelligence•10.1IntelligenceandMachines•10.2Perception•10.3Reasoning•10.4AdditionalAreasofResearch•10.5ArtificialNeuralNetworks•10.6Robotics•10.7ConsideringtheConsequences©2007PearsonAddison-Wesley.Allrightsreserved0-3IntelligentAgents•Agent:A“device”thatrespondstostimulifromitsenvironment–Sensors–Actuators•Thegoalofartificialintelligenceistobuildagentsthatbehaveintelligently©2007PearsonAddison-Wesley.Allrightsreserved0-4LevelsofIntelligentBehavior•Reflex:actionsarepredeterminedresponsestotheinputdata•Intelligentresponse:actionsaffectedbyknowledgeoftheenvironment•Goalseeking•Learning©2007PearsonAddison-Wesley.Allrightsreserved0-5Figure10.1Theeight-puzzleinitssolvedconfiguration©2007PearsonAddison-Wesley.Allrightsreserved0-6Figure10.2Ourpuzzle-solvingmachine©2007PearsonAddison-Wesley.Allrightsreserved0-7ApproachestoResearchinArtificialIntelligence•Performanceoriented:Researchertriestomaximizetheperformanceoftheagents.•Simulationoriented:Researchertriestounderstandhowtheagentsproduceresponses.©2007PearsonAddison-Wesley.Allrightsreserved0-8TuringTest•ProposedbyAlanTuringin1950•Benchmarkforprogressinartificialintelligence•Testsetup:Humaninterrogatorcommunicateswithtestsubjectbytypewriter.•Test:Canthehumaninterrogatordistinguishwhetherthetestsubjectishumanormachine?©2007PearsonAddison-Wesley.Allrightsreserved0-9TechniquesforUnderstandingImages•Templatematching•Imageprocessing–edgeenhancement–regionfinding–smoothing•Imageanalysis©2007PearsonAddison-Wesley.Allrightsreserved0-10LanguageProcessing•SyntacticAnalysis•SemanticAnalysis•ContextualAnalysis©2007PearsonAddison-Wesley.Allrightsreserved0-11Figure10.3Asemanticnet©2007PearsonAddison-Wesley.Allrightsreserved0-12ComponentsofaProductionSystems1.Collectionofstates–Start(orinitial)state–Goalstate(orstates)2.Collectionofproductions:rulesormoves–Eachproductionmayhavepreconditions3.Controlsystem:decideswhichproductiontoapplynext©2007PearsonAddison-Wesley.Allrightsreserved0-13ReasoningbySearching•StateGraph:Allstatesandproductions•SearchTree:Arecordofstatetransitionsexploredwhilesearchingforagoalstate–Breadth-firstsearch–Depth-firstsearch©2007PearsonAddison-Wesley.Allrightsreserved0-14Figure10.4Asmallportionoftheeight-puzzle’sstategraph©2007PearsonAddison-Wesley.Allrightsreserved0-15Figure10.5Deductivereasoninginthecontextofaproductionsystem©2007PearsonAddison-Wesley.Allrightsreserved0-16Figure10.6Anunsolvedeight-puzzle©2007PearsonAddison-Wesley.Allrightsreserved0-17Figure10.7Asamplesearchtree©2007PearsonAddison-Wesley.Allrightsreserved0-18Figure10.8Productionsstackedforlaterexecution©2007PearsonAddison-Wesley.Allrightsreserved0-19HeuristicStrategies•Heuristic:Aquantitativeestimateofthedistancetoagoal•Requirementsforgoodheuristics–Mustbemucheasiertocomputethanacompletesolution–Mustprovideareasonableestimateofproximitytoagoal©2007PearsonAddison-Wesley.Allrightsreserved0-20Figure10.9Anunsolvedeight-puzzle©2007PearsonAddison-Wesley.Allrightsreserved0-21Figure10.10Analgorithmforacontrolsystemusingheuristics©2007PearsonAddison-Wesley.Allrightsreserved0-22Figure10.11Thebeginningsofourheuristicsearch©2007PearsonAddison-Wesley.Allrightsreserved0-23Figure10.12Thesearchtreeaftertwopasses©2007PearsonAddison-Wesley.Allrightsreserved0-24Figure10.13ThesearchtreeafterthreepassesFigure10.14Thecompletesearchtreeformedbyourheuristicsystem©2007PearsonAddison-Wesley.Allrightsreserved0-26HandlingReal-WorldKnowledge•Representationandstorage•Accessingrelevantinformation–Meta-Reasoning–Closed-WorldAssumption•Frameproblem©2007PearsonAddison-Wesley.Allrightsreserved0-27Learning•Imitation•SupervisedTraining•Reinforcement•EvolutionaryTechniques©2007PearsonAddison-Wesley.Allrightsreserved0-28ArtificialNeuralNetworks•ArtificialNeuron–Eachinputismultipliedbyaweightingfactor.–Outputis1ifsumofweightedinputsexceedsthethresholdvalue;0otherwise.•Networkisprogrammedbyadjustingweightsusingfeedbackfromexamples.©2007PearsonAddison-Wesley.Allrightsreserved0-29Figure10.15Aneuroninalivingbiologicalsystem©2007PearsonAddison-Wesley.Allrightsreserved0-30Figure10.16Theactivitieswithinaprocessingunit©2007PearsonAddison-Wesley.Allrightsreserved0-31Figure10.17Representationofaprocessingunit©2007PearsonAddison-Wesley.Allrightsreserved0-32Figure10.18Aneuralnetworkwithtwodifferentprograms©2007PearsonAddison-Wesley.Allrightsreserved0-33Figure10.19Anartificialneuralnetwork©2007PearsonAddison-Wesley.Allrightsreserved0-34Figure10.20Traininganartificialneuralnetwork©2007PearsonAddison-Wesley.Allrightsreserved0-35Figure10.20Traininganartificialneuralnetwork(continued)©2007PearsonAddison-Wesley.Allrightsreserved0-36Figure10.20Traininganartificialneuralnetwork(continued)©2007PearsonAddison-Wesley.Allrightsreserved0-37Figure10.20Traininganartificialneuralnetwork(continued)©2007PearsonAddison-Wesley.Allrightsreserved0-38Figure10.21ThestructureofALVINN©2007PearsonAddison-Wesley.Allrightsreser