csc 4510 – machine learningmap/4510/02mlhistory-s-u.pdf · 2012-01-26 · machine learning •...
Post on 22-May-2020
3 Views
Preview:
TRANSCRIPT
CSC4510–MachineLearningDr.Mary‐AngelaPapalaskari
DepartmentofCompuBngSciencesVillanovaUniversity
Coursewebsite:
www.csc.villanova.edu/~map/4510/
Lecture2:HistoryandOverviewofMachineLearning
1CSC4510‐M.A.Papalaskari‐VillanovaUniversity
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 2
“Itwon’ttrulybeanautonomousvehicleunBlyouinstructittodrivetoworkanditheadstothebeachinstead.”
‐ BradTempleton,SoTwaredesignerandaconsultantfortheGoogleprojectonAutonomousVehicles
‐ NYTimes1/24/12‐ hYp://www.nyBmes.com/2012/01/24/technology/googles‐autonomous‐vehicles‐draw‐skepBcism‐at‐legal‐symposium.html?_r=2&nl=technology&emc=techupdateema22
WhatarethegoalsofAIresearch?
ArBfactsthatACTlikeHUMANS
ArBfactsthatTHINKlikeHUMANS
ArBfactsthatTHINKRATIONALLY
ArBfactsthatACTRATIONALLY
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 3
ABitofHistory
• ArthurSamuel(1959)wroteaprogramthatlearnttoplaycheckerswellenoughtobeathim.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 4
1940sAdvancesinmathemaBcallogic,informaBontheory,conceptofneuralcomputaBon
1943:McCulloch&PiYsNeuron 1948:Shannon:InformaBonTheory 1949:HebbianLearning cellsthatfiretogether,wiretogether
1950sEarlycomputers.Dartmouthconferencecoinsthephrase“arBficialintelligence”andLispisproposedastheAIprogramminglanguage
1950:TuringTest 1956:DartmouthConference 1958:Friedberg:LearnAssemblyCode 1959:Samuel:LearningCheckers
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 5
1960sA.I.fundingincreased(mainlymilitary).Famousquote:“WithinageneraBon...theproblemofcreaBng'arBficialintelligence'willsubstanBallybesolved.”Earlysymbolicreasoningapproaches.
LogicTheorist,GPS,Perceptrons 1969: Minsky & Papert “Perceptrons”
1970sA.I.“winter”–Fundingdriesupaspeoplerealizethisisahardproblem!LimitedcompuBngpoweranddead‐endframeworksleadtofailures.
eg:MachineTranslaBonFailure
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 6
1980sRulebased“expertsystems”usedinmedical/legalprofessions.Bio‐inspiredalgorithms(Neuralnetworks,GeneBcAlgorithms).Again:A.I.promisestheworld–lotsofcommercialinvestment
ExpertSystems(Mycin,Dendral,EMYCINKnowledgeRepresentaBonandreasoning:
Frames,Eurisko,Cyc,NMR,fuzzylogicSpeechRecogniBon(HEARSAY,HARPY,HWIM)
ML: 1982: Hopfield Nets, Decision Trees, GA & GP. 1986: Backpropagation, Explanation-Based Learning
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 7
1990sSomeconcretesuccessesbegintoemerge.AIdivergesintoseparatefields:ComputerVision,AutomatedReasoning,Planningsystems,NaturalLanguageprocessing,MachineLearning…
…MachineLearningbeginstooverlapwithsta4s4cs/probabilitytheory.
1992: Koza & Genetic Programming 1995: Vapnik: Support Vector Machines
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 8
2000s
Firstcommercial‐strengthapplicaBons:Google,Amazon,computergames,route‐finding,creditcardfrauddetecBon,spamfilters,etc…
Toolsadoptedasstandardbyotherfieldse.g.biology
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 9
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 10
2010s…. ??????
• Usingmachinelearningtodetectspamemails.
To: you@gmail.com GET YOUR DIPLOMA TODAY! If you are looking for a fast and cheap way to get a diploma, this is the best way out for you. Choose the desired field and degree and call us right now: For US: 1.845.709.8044 Outside US: +1.845.709.8044 "Just leave your NAME & PHONE NO. (with CountryCode)" in the voicemail. Our staff will get back to you in next few days!
ALGORITHM Naïve Bayes Rule mining
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 11
• Usingmachinelearningtorecommendbooks.
ALGORITHMS Collaborative Filtering Nearest Neighbour Clustering CSC4510‐M.A.Papalaskari‐VillanovaUniversity 12
• UsingmachinelearningtoidenBfyfacesandexpressions.
ALGORITHMS Decision Trees
Adaboost
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 13
ALGORITHMS Feature Extraction Probabilistic Classifiers Support Vector Machines + many more….
• Using machine learning to identify vocal patterns
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 14
• MLforworkingwithsocialnetworkdata:detecBngfraud,predicBngclick‐thrupaYerns,targetedadverBsing,etcetcetc.
ALGORITHMS Support Vector Machines Collaborative filtering Rule mining algorithms Many many more….
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 15
• ArthurSamuel(1959).MachineLearning:Fieldofstudythatgivescomputerstheabilitytolearnwithoutbeingexplicitlyprogrammed.
Samuel’sdefiniBonofMLissBllrelevant
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 16
AcomputerprogramissaidtolearnfromexperienceEwithrespecttosometaskTandsomeperformancemeasureP,ifitsperformanceonT,asmeasuredbyP,improveswithexperienceE.
TomMitchell(1998):Well‐posedLearningProblem
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 17
DefiningtheLearningTaskImproveontask,T,withrespectto
performancemetric,P,basedonexperience,E.T:PlayingcheckersP:PercentageofgameswonagainstanarbitraryopponentE:PlayingpracBcegamesagainstitself
T:Recognizinghand‐wriYenwordsP:PercentageofwordscorrectlyclassifiedE:Databaseofhuman‐labeledimagesofhandwriYenwords
T:Drivingonfour‐lanehighwaysusingvisionsensorsP:Averagedistancetraveledbeforeahuman‐judgederrorE:Asequenceofimagesandsteeringcommandsrecordedwhileobservingahumandriver.
T:DeterminewhichstudentslikeorangesorapplesP:Percentageofstudents’preferencesguessedcorrectlyE:StudentaYributedata
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 18
DesigningaLearningSystem• Choosethetrainingexperience• Chooseexactlywhatistoobelearned,i.e.thetargetfunc>on.• ChoosealearningalgorithmtoinferthetargetfuncBonfromthe
experience.• Alearningalgorithmwillalsodetermineaperformancemeasure
Environment/Experience
Learner
Knowledge
PerformanceElement
CSC4510‐M.A.Papalaskari‐VillanovaUniversity
19
Improveontask,T,withrespecttoperformancemetric,P,basedonexperience,E.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Supposeyouremailprogramwatcheswhichemailsyoudoordonotmarkasspam,andbasedonthatlearnshowtobeYerfilterspam.WhatisthetaskTinthisseAng?
• Watchingyoulabelemailsasspamornotspam.• Classifyingemailsasspamornotspam• Thenumber(orfracBon)ofemailscorrectlyclassifiedasspam/notspam.• Noneoftheabove—thisisnotamachinelearningproblem.
20
Quick check:
Machinelearning• SupervisedLearning
– ClassificaBon– Regression
• Unsupervisedlearning
Others:Reinforcementlearning,recommendersystems.
Alsotalkabout:PracBcaladviceforapplyinglearningalgorithms.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 21
Machinelearning• SupervisedLearning
– ClassificaBon– Regression
• Unsupervisedlearning
Others:Reinforcementlearning,recommendersystems.
Alsotalkabout:PracBcaladviceforapplyinglearningalgorithms.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 22
ClassificaBon
• Example:Creditscoring
• DifferenBaBngbetweenlow‐riskandhigh‐riskcustomersfromtheirincomeandsavings
Discriminant:IFincome>θ1ANDsavings>θ2 THENlow‐riskELSEhigh‐risk
23CSC4510‐M.A.Papalaskari‐VillanovaUniversity
ClassificaBon• Example:Irisdata• 4aYributes
– sepallength– sepalwidth– petallength– petalwidth
• DifferenBaBngbetween3differenttypesofiris
24CSC4510‐M.A.Papalaskari‐VillanovaUniversity
25CSC4510‐M.A.Papalaskari‐VillanovaUniversity
IrisDatamoreplots:
ClassificaBonTree
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 26
FaceRecogniBon
Training examples of a person
Test images
ORL dataset, AT&T Laboratories, Cambridge UK
27CSC4510‐M.A.Papalaskari‐VillanovaUniversity
0
100
200
300
400
0 500 1000 1500 2000 2500
HousingpricepredicBon.
Price ($) in 1000’s
Size in feet2
Regression:PredictconBnuousvaluedoutput(price)
SupervisedLearning
“rightanswers”given
28CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Machinelearning• SupervisedLearning
– ClassificaBon– Regression
• Unsupervisedlearning
Others:Reinforcementlearning,recommendersystems.
Alsotalkabout:PracBcaladviceforapplyinglearningalgorithms.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 29
Regression
• Example:Priceofausedcar
• x:caraYributesy:price
y=g(x|θ )
g()model,
θ parameters
y=wx+w0
30
CSC4510‐M.A.Papalaskari‐VillanovaUniversity
RegressionApplicaBons
• NavigaBngacar:Angleofthesteering• KinemaBcsofarobotarm
31CSC4510‐M.A.Papalaskari‐VillanovaUniversity
SupervisedLearning:Uses• PredicBonoffuturecases:Usetheruletopredicttheoutputforfutureinputs
• KnowledgeextracBon:Theruleiseasytounderstand
• Compression:Theruleissimplerthanthedataitexplains
• OutlierdetecBon:ExcepBonsthatarenotcoveredbytherule,e.g.,fraud
32CSC4510‐M.A.Papalaskari‐VillanovaUniversity
• Treat both as classification problems.
• Treat problem 1 as a classification problem, problem 2 as a regression problem. • Treat problem 1 as a regression problem, problem 2 as a classification problem. • Treat both as regression problems.
You’re running a company, and you want to develop learning algorithms to address each of two problems.
Problem 1: You have a large inventory of identical items. You want to predict how many of these items will sell over the next 3 months. Problem 2: You’d like software to examine individual customer accounts, and for each account decide if it has been hacked/compromised.
Should you treat these as classification or as regression problems?
Quick check:
Machinelearning• SupervisedLearning
– ClassificaBon– Regression
• Unsupervisedlearning
Others:Reinforcementlearning,recommendersystems.
Alsotalkabout:PracBcaladviceforapplyinglearningalgorithms.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 34
x1
x2
Supervised Learning
35CSC4510‐M.A.Papalaskari‐VillanovaUniversity
x1
x2
Unsupervised Learning
36CSC4510‐M.A.Papalaskari‐VillanovaUniversity
UnsupervisedLearning
• Learning“whatnormallyhappens”• Nooutput• Clustering:Groupingsimilarinstances
• ExampleapplicaBons– CustomersegmentaBon– Imagecompression:ColorquanBzaBon
– BioinformaBcs:LearningmoBfs
37CSC4510‐M.A.Papalaskari‐VillanovaUniversity
38CSC4510‐M.A.Papalaskari‐VillanovaUniversity
39CSC4510‐M.A.Papalaskari‐VillanovaUniversity
40CSC4510‐M.A.Papalaskari‐VillanovaUniversity
41CSC4510‐M.A.Papalaskari‐VillanovaUniversity
[Source: Su-In Lee, Dana Pe’er, Aimee Dudley, George Church, Daphne Koller]
Gen
es
Individuals
42CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Organize computing clusters Social network analysis
Image credit: NASA/JPL-Caltech/E. Churchwell (Univ. of Wisconsin, Madison)
Astronomical data analysis Market segmentation 43CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Ofthefollowingexamples,whichwouldyouaddressusinganunsupervisedlearningalgorithm?(Checkallthatapply.)
Given a database of customer data, automatically discover market segments and group customers into different market segments.
• Given email labeled as spam/not spam, learn a spam filter.
• Given a set of web pages found on the web, automatically detect the ones that are syllabi for AI or software engineering courses • Given a dataset of patients diagnosed as either having diabetes or not, learn to classify new patients as having diabetes or not.
Quick check:
• Given a database of nutrition data, automatically discover categories of food items.
Machinelearning• SupervisedLearning
– ClassificaBon– Regression
• Unsupervisedlearning
Others:Reinforcementlearning,recommendersystems.
Alsotalkabout:PracBcaladviceforapplyinglearningalgorithms.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 45
ReinforcementLearning
• Learningapolicy:Asequenceofoutputs• Nosupervisedoutputbutdelayedreward• Creditassignmentproblem
• Gameplaying
• Robotinamaze
• MulBpleagents,parBalobservability,...
46CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Machinelearning• SupervisedLearning
– ClassificaBon– Regression
• Unsupervisedlearning
Others:Reinforcementlearning,recommendersystems.
Alsotalkabout:PracBcaladviceforapplyinglearningalgorithms.
CSC4510‐M.A.Papalaskari‐VillanovaUniversity 47
Supervised or Unsupervised learning? Iris Data
Summary• MLgrewoutofworkinAI
Op>mizeaperformancecriterionusingexampledataorpastexperience.
• Typesoflearning– Supervised– Unsupervised
• RoleofStaBsBcs:Inferencefromasample
• RoleofComputerscience:– DatarepresentaBonandmodeling
– EfficientalgorithmstosolveopBmizaBonproblems– RepresenBngandevaluaBngthemodelforinference
49CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Resources:Datasets
• UCIRepository:hYp://www.ics.uci.edu/~mlearn/MLRepository.html
• UCIKDDArchive:hYp://kdd.ics.uci.edu/summary.data.applicaBon.html
• Statlib:hYp://lib.stat.cmu.edu/
• Delve:hYp://www.cs.utoronto.ca/~delve/
50CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Resources:Journals
• JournalofMachineLearningResearchwww.jmlr.org• MachineLearning• NeuralComputaBon• NeuralNetworks• IEEETransacBonsonNeuralNetworks• IEEETransacBonsonPaYernAnalysisandMachineIntelligence
• AnnalsofStaBsBcs• JournaloftheAmericanStaBsBcalAssociaBon• ...
51CSC4510‐M.A.Papalaskari‐VillanovaUniversity
Resources:Conferences
• InternaBonalConferenceonMachineLearning(ICML)• EuropeanConferenceonMachineLearning(ECML)• NeuralInformaBonProcessingSystems(NIPS)• UncertaintyinArBficialIntelligence(UAI)• ComputaBonalLearningTheory(COLT)• InternaBonalConferenceonArBficialNeuralNetworks
(ICANN)• InternaBonalConferenceonAI&StaBsBcs(AISTATS)• InternaBonalConferenceonPaYernRecogniBon(ICPR)• ...
52CSC4510‐M.A.Papalaskari‐VillanovaUniversity
SomeoftheslidesinthispresentaBonareadaptedfrom:
• Prof.FrankKlassner’sMLclassatVillanova• theUniversityofManchesterMLcoursehYp://www.cs.manchester.ac.uk/ugt/COMP24111/
• TheStanfordonlineMLcoursehYp://www.ml‐class.org/
top related