斯坦福大学机器学习第一讲Lecture1.docx

斯坦福大学机器学习第一讲Lecture1.docx

《斯坦福大学机器学习第一讲Lecture1.docx》由会员分享，可在线阅读，更多相关《斯坦福大学机器学习第一讲Lecture1.docx（19页珍藏版）》请在冰豆网上搜索。

斯坦福大学机器学习第一讲Lecture1

Introduction

Welcome

MachineLearning

MachineLearning

-GrewoutofworkinAI

-Newcapabilityforcomputers

Examples:

-Databasemining

Largedatasetsfromgrowthofautomation/web.

E.g.,Webclickdata,medicalrecords,biology,engineering-Applicationscan’tprogrambyhand.

E.g.,Autonomoushelicopter,handwritingrecognition,mostofNaturalLanguageProcessing（NLP,ComputerVision.

-Newcapabilityforcomputers

Examples:

-Databasemining

Largedatasetsfromgrowthofautomation/web.

E.g.,Webclickdata,medicalrecords,biology,engineering-Applicationscan’tprogrambyhand.

E.g.,Autonomoushelicopter,handwritingrecognition,mostofNaturalLanguageProcessing（NLP,ComputerVision.

-Newcapabilityforcomputers

Examples:

-Databasemining

Largedatasetsfromgrowthofautomation/web.

E.g.,Webclickdata,medicalrecords,biology,engineering-Applicationscan’tprogrambyhand.

E.g.,Autonomoushelicopter,handwritingrecognition,mostofNaturalLanguageProcessing（NLP,ComputerVision.

-Newcapabilityforcomputers

Examples:

-Databasemining

Largedatasetsfromgrowthofautomation/web.

E.g.,Webclickdata,medicalrecords,biology,engineering-Applicationscan’tprogrambyhand.

E.g.,Autonomoushelicopter,handwritingrecognition,mostofNaturalLanguageProcessing（NLP,ComputerVision.-Self-customizingprograms

E.g.,Amazon,Netflixproductrecommendations

-Newcapabilityforcomputers

Examples:

-Databasemining

Largedatasetsfromgrowthofautomation/web.

E.g.,Webclickdata,medicalrecords,biology,engineering-Applicationscan’tprogrambyhand.

E.g.,Autonomoushelicopter,handwritingrecognition,mostofNaturalLanguageProcessing（NLP,ComputerVision.-Self-customizingprograms

E.g.,Amazon,Netflixproductrecommendations

Introduction

Whatismachine

learning

MachineLearning

•ArthurSamuel（1959.MachineLearning:

Fieldofstudythatgivescomputerstheabilitytolearnwithoutbeingexplicitlyprogrammed.

•ArthurSamuel（1959.MachineLearning:

Fieldofstudythatgivescomputerstheabilitytolearnwithoutbeingexplicitlyprogrammed.

MachineLearningdefinition

•ArthurSamuel（1959.MachineLearning:

Fieldofstudythatgivescomputerstheabilitytolearnwithoutbeingexplicitlyprogrammed.

•TomMitchell（1998Well-posedLearningProblem:

AcomputerprogramissaidtolearnfromexperienceEwithrespecttosometaskTandsomeperformancemeasureP,ifits

performanceonT,asmeasuredbyP,improveswithexperienceE.

MachineLearningdefinition

Classifyingemailsasspamornotspam.Thenumber（orfractionofemailscorrectlyclassifiedasspam/notspam.Noneoftheabove—thisisnotamachinelearningproblem.Supposeyouremailprogramwatcheswhichemailsyoudoordonotmarkasspam,andbasedonthatlearnshowtobetterfilterspam.WhatisthetaskTinthissetting?

“AcomputerprogramissaidtolearnfromexperienceEwithrespecttosometaskTandsomeperformancemeasureP,ifitsperformanceonT,asmeasuredbyP,improveswithexperienceE.”

Classifyingemailsasspamornotspam.Supposeyouremailprogramwatcheswhichemailsyoudoordonotmarkasspam,andbasedonthatlearnshowtobetterfilterspam.WhatisthetaskTinthissetting?

“AcomputerprogramissaidtolearnfromexperienceEwithrespecttosometaskTandsomeperformancemeasureP,ifitsperformanceonT,asmeasuredbyP,improveswithexperienceE.”

Classifyingemailsasspamornotspam.Thenumber（orfractionofemailscorrectlyclassifiedasspam/notspam.Noneoftheabove—thisisnotamachinelearningproblem.Supposeyouremailprogramwatcheswhichemailsyoudoordonotmarkasspam,andbasedonthatlearnshowtobetterfilterspam.WhatisthetaskTinthissetting?

“AcomputerprogramissaidtolearnfromexperienceEwithrespecttosometaskTandsomeperformancemeasureP,ifitsperformanceonT,asmeasuredbyP,improveswithexperienceE.”

Machinelearningalgorithms:

-Supervisedlearning

-UnsupervisedlearningAlsotalkabout:

Practicaladviceforapplyinglearningalgorithms.

Introduction

Supervised

Learning

MachineLearning

100

200300

4000500

150********0

Housingpriceprediction.

Price（$in1000’s

Sizeinfeet2

Predictcontinuous

Breastcancer（malignant,benign

Discretevaluedoutput（0or1

Malignant?

1（Y0（N

Age-ClumpThickness--UniformityofCellShape…

Treatbothasclassificationproblems.

Treatproblem1asaclassificationproblem,problem2asaregressionproblem.Treatproblem1asaregressionproblem,problem2asaclassificationproblem.Treatbothasregressionproblems.You’rerunningacompany,andyouwanttodeveloplearningalgorithmstoaddresseachoftwoproblems.Problem1:

Youhavealargeinventoryofidenticalitems.Youwanttopredicthowmanyoftheseitemswillselloverthenext3months.Problem2:

You’dlikesoftwaretoexamineindividualcustomeraccounts,andforeachaccountdecideifithasbeenhacked/compromised.

AndrewNg

AndrewNgIntroductionUnsupervised

LearningMachineLearning

AndrewNgx1

x2

SupervisedLearning

AndrewNg

UnsupervisedLearningx1

x2

AndrewNg

Genes

Individuals

Genes

Individuals

OrganizecomputingclustersImagecredit:

NASA/JPL-Caltech/E.Churchwell（Univ.ofWisconsin,Madison

Cocktailpartyproblem

Microphone#1Microphone#2

Speaker#1Speaker#2

Microphone#1:

Microphone#2:

Microphone#1:

Microphone#2:

[AudioclipscourtesyofTe-WonLee.]Output#1:

Output#2:

Output#1:

Output#2:

AndrewNg

Cocktailpartyproblemalgorithm[W,s,v]=svd（（repmat（sum（x.*x,1,size（x,1,1.*x*x';[Source:

SamRoweis,YairWeiss&EeroSimoncelli]AndrewNg

Ofthefollowingexamples,whichwouldyouaddressusinganunsupervisedlearningalgorithm?

（Checkallthatapply.Givenemaillabeledasspam/notspam,learnaspamfilter.Givenasetofnewsarticlesfoundontheweb,groupthemintosetofarticlesaboutthesamestory.Givenadatabaseofcustomerdata,automaticallydiscovermarketsegmentsandgroupcustomersintodifferentmarketsegments.Givenadatasetofpatientsdiagnosedaseitherhavingdiabetesornot,learntoclassifynewpatientsashavingdiabetesornot.

AndrewNg