《人工智能详解课件.ppt》由会员分享,可在线阅读,更多相关《人工智能详解课件.ppt(69页珍藏版)》请在得力文库 - 分享文档赚钱的网站上搜索。
1、Modern Artificial Intelligence and Its Importance in the Future WorldZengchang Qin(Ph.D.)Intelligent Computing and Machine Learning LabSchool of Automation and Electrical EngineeringBeihang UniversityShahe Campus Oct 27 2010This is ScienceGiveabigpictureofmodernArtificialIntelligenceandunderstandwhy
2、itisimportantinthecurrentandthefutureworld.WehavesuchadirectionofresearchintheschoolofASEE.ToclarifythemisunderstandingofA.I.fromthoserobotmoviesandsciencefictions.About This TalkIhavebeenworkinginA.I.areforthepastdecade.Ienjoymoviesandunboundedthinking.Iamalwaysintriguedbyanykindsfexcellentideasfro
3、mhumanintelligence.Feelfreetoaskanyquestionsyouhaveinmind,noguaranteetobeanswered.About The SpeakerMisunderstandingArtificialIntelligence(A.I.)RoboticsJohnMcCarthy(Stanford)Artificial Intelligence We fear?I,RobotTheThreeLawsofRoboticsbyIssacAsimovareasthefollows:Arobotmaynotinjureahumanbeingor,throu
4、ghinaction,allowahumanbeingtocometoharm.Arobotmustobeyanyordersgiventoitbyhumanbeings,exceptwheresuchorderswouldconflictwiththeFirstLaw.ArobotmustprotectitsownexistenceaslongassuchprotectiondoesnotconflictwiththeFirstorSecondLaw.My Philosophy of Modern A.I.ArtificialIntelligenceisamathematical/compu
5、tingtechnologythatwillmakelifebetter.Ihavebeeninterestedinmakingmachinesintelligentbydesigningalgorithms.Imaynotbelievethatonedaywecanrecreatehumanbrainsusingsiliconchips,butIbelievethatcomputingwillaidourbrainstodomissionsimpossibleinthefuture.Chinese Room ParadoxModern A.I.The Engineering Approach
6、:Machine Learning and Data MiningPatternRecognition,ComputervisionandImageProcessingDistributedA.I./multi-agentsystemsBiometricsandcomputerforensicsNaturalLanguageProcessingIntelligentSearchandInformationRetrievalComputationalCognitiveScienceComputationalNeuroscienceandbioinformaticsComputationalCog
7、nitiveScienceComputational/BehaviorFinanceBehaviorTargetingandPersonalServicesDigitalAdvertisements/recommendationsystemsPhilosophy of Machine LearningMachineLearningsearchinthehypothesisspacetofindtheonesthatmatchthedata.UsingOccamsrazor,wechoosethesimplestone.WilliamofOckham(orOccam)wasa14th-centu
8、ryEnglishlogicianandFranciscanfriarwhosnameisgiventotheprinciplethatwhentryingtochoosebetweenmultiplecompetingtheoriesthesimplesttheoryisprobablythebest.ThisprincipleisknownasOckhamsrazor.ExampleExample 2Why Machine Learning is important?Tofinethetheorythatexplainsthedata,weusuallypreferthesimpleone
9、s.Machinelearningandscientificdiscoverysharesimilarities.KarlPopperLogic ProgrammingLondonUndergroundExampleFuzzy LogicMembership function(continuous)Membership FunctionsSome Intuition Professor of Fuzzy Logic Multi-agent SystemDistributedA.I.-coordinationDataminingistheprocessofextractingpatternsfr
10、omdata-Torturethedatauntiltheyconfess.Dataiseverywhereandindifferenttypes.Pattern Recognition and Data MiningWelcome to FairmontNET.stdtext font-family:Verdana,Arial,Helvetica,sans-serif;font-size:11px;color:#1F3D4E;.stdtext_wh font-family:Verdana,Arial,Helvetica,sans-serif;font-size:11px;color:WHIT
11、E; HTML and EmailsReturn-pathReceivedfromrelay2.EECS.Berkeley.EDU(relay2.EECS.Berkeley.EDU169.229.60.28)byimap4.CS.Berkeley.EDU(iPlanetMessagingServer5.2HotFix1.16(builtMay142003)withESMTPid;Tue,08Jun200411:40:43-0700(PDT)Receivedfromrelay3.EECS.Berkeley.EDU(localhost127.0.0.1)byrelay2.EE
12、CS.Berkeley.EDU(8.12.10/8.9.3)withESMTPidi58Ieg3N000927;Tue,08Jun200411:40:43-0700(PDT)Receivedfromredbirds(dhcp-168-35.EECS.Berkeley.EDU128.32.168.35)byrelay3.EECS.Berkeley.EDU(8.12.10/8.9.3)withESMTPidi58IegFp007613;Tue,08Jun200411:40:42-0700(PDT)DateTue,08Jun200411:40:42-0700FromRobertMillerSubje
13、ctRE:SLTheadcount=25In-reply-toToRandyKatzCcGlendaJ.Smith,GertLanckrietMessage-idMIME-version1.0X-MIMEOLEProducedByMicrosoftMimeOLEV6.00.2800.1409X-MailerMicrosoftOfficeOutlook,Build11.0.5510Content-typemultipart/alternative;boundary=-=_NextPart_000_0033_01C44D4D.6DD93AF0Thread-indexAcRMtQRp+R26lVFa
14、Riuz4BfImikTRAA0wf3Qtheheadcountisnow32.-RobertMiller,AdministrativeSpecialistUniversityofCalifornia,BerkeleyElectronicsResearchLab634SodaHall#1776Berkeley,CA94720-1776Phone:510-642-6037fax:510-643-128924Medical Image,handwritten recognition25Sounds-fingerprints26Intelligent Search and Bio-identityM
15、irco-array Data of GenesDrug DesignsComputer Human Interface EEG signalsStock IndexData Types fraud detectionSocial Network MiningMonitoringfluthroughtwitter.Monitoringtrafficthroughmobilecalls.Entity Cube34Experimental EconomicsVernonL.Smithforhavingestablishedlaboratoryexperimentsasatoolinempirica
16、leconomicanalysis,especiallyinthestudyofalternativemarketmechanisms”From http:/nobelprize.org/Behavior Economics Irrational AgentsNotableforhisworkonthepsychologyofjudgmentsanddecisionmaking,behavioraleconomics.Winning$10or$1000withchanceof1%.Losing$10or$1000withchanceof1%Software Agents for Trading
17、WhatisthecapitalofChina?WhatisthepopulationofBeijing?WhatisthepopulationofthecapitalofChina?Reasoning with Natural Language Evolutionary ComputingGeneticAlgorithmSirRichardDawkins“TheselfishGenes”Stochastic OptimizationCellular AutomatonWolframwaseducatedatEton.Attheageof15,hepublishedanarticleonpar
18、ticlephysics4andenteredOxfordUniversityatage17.Hewroteawidelycitedpaperonheavyquarkproductionatage18.2WolframreceivedhisPh.D.inparticlephysicsfromtheCaliforniaInstituteofTechnologyatage205andjoinedthefacultythere.Hebecamehighlyinterestedincellularautomataatage21.2Wolframsworkinparticlephysics,cosmol
19、ogyandcomputerscienceearnedhimoneofthefirstMacArthurawards.Decision TreesP(h|e)=P(e|h)P(h)/P(e)AProofthateveryonecanunderstandP(h,e)=P(h|e)P(e)P(e,h)=P(e|h)P(h)Bayesian StatisticsGraphical Model of Gaussian Distribution and Hiearachical Structure with Latent VariablesUnderstanding SemanticsDemograph
20、ics MS AdCenter LabCommercial Intentions of Given WebsiteIfyouwanttosellone,whatisthebestprice?N97(Nokia Phone)Minority GameEIFarolBarMinorityGameModelApplicationInRealworldTherearemorethan100IrishmusicloversbutElFarolhasonly60seats.Theshowisenjoyableonlywhenfewerthan60peopleshowup.Everypeopleshould
21、decideweeklywhethergotothebartoenjoylivemusicintheriskofstayinginacrowdplaceorstayathome.Therulesaresimple:afinitenumberofplayershavetochoosebetweentwosides;whoeverendsupintheminoritysideisawinner.SimplifiedfrommarketaimingtoanalyzecomplexfinancialmarketCollective Behavior DecompositionSimulation Re
22、sults(Li,Ma and Qin,2010)YingMa,GuanyiLi,YingsaiDongandZengchang Qin(2010),Minoritygamedataminingformarketpredictions,forStockMarketPredictions,toappearintheProceedingsofAAMAS2010.GuanyiLi,YingMa,YingsaiDongandZengchang Qin(2010),Behaviorlearninginminoritygames,ToappearintheProceedingsofCARE2009.Zen
23、gchang Qin,MarcusThintandZhihengHuang(2009),Rankinganswersbyhierarchicaltopicmodels,ProceedingsofIEA/AIE2009,LNCS5579,pp.103-112,Springer.ZhihengHuang,MarcusThintandZengchang Qin(2008),Questionclassificationusingheadwordsandtheirhypernyms,TheProceedingsofConferenceonEmpiricalMethodsonNaturalLanguageProcessing,pp.927-936,ACL.ReferencesNon-academicAcademic AIFuzzy Logic and Logic of ScienceNLP&ANNGA,ALIFE&Multi-agentWeb:orGoogle“ZengchangQin”formyLinkedInProfiles.Contact InformationThankyouverymuch!Anyquestions?
限制150内