CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine...

13
CSE 517A – MACHINE LEARNING Spring 2019 Marion Neumann COURSE OVERVIEW & STRUCTURE

Transcript of CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine...

Page 1: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

CSE517A– MACHINELEARNING

Spring2019MarionNeumann

COURSEOVERVIEW&STRUCTURE

Page 2: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

ABOUT

• MarionNeumann• office:JolleyHall222• officehours:TUE 11:30-12:30pm• contact:usePiazza(http://piazza.com/wustl/spring2019/cse517a)

• Lectures:TUE&THU• 10-11:30pmin Louderman458

• Coursewebsite: https://sites.wustl.edu/neumann/courses/cse517a/sp19/

2

Bookmarkme!!!

Youareareal person!

Page 3: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

READING

CourseBook• AFirstCourseinMachineLearning,

RogersandGirolami,2nded.(Wewillusethisbookforreadings,mathematicalderivations, andhomework problems.)

Allreadingwillbepostedonthecoursewebpageandisconsideredcoursematerialsand exam-relevant!

3

Resources:https://sites.wustl.edu/neumann/courses/cse517a/resources/

Getacopyofthisbook!

Page 4: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

GRADINGANDPOLICIES

4

CourseSyllabus:https://sites.wustl.edu/neumann/courses/cse517a/syllabus/

Bystayingenrolledinthiscourseyouconfirmthatyouread,understood and agreed

thecoursesyllabus.

Page 5: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

COLLABORATIONPOLICY

Collaboration:yes• discusscoursematerialswithotherstudentsà joinastudygroup

Cheating:no• donotcopyanswers/codeorpartsofanswers/codefromanyoneelseorfromanymaterialyoufindonline

5

AutogradercomparesyoursolutionstothesolutionsofotherstudentsandsolutionsavailableONLINE!

TAswillactivelywatchoutforsimilarsolutionstowrittenhws!

Page 6: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

ACADEMICINTEGRITY

• Everythingthatyouturninforthiscoursemustbeyourownwork.Ifyouwillfullymisrepresentsomeoneelse’sworkasyourown,youareguiltyofcheating.

• Providing yourcoursework(writtenorcode)in anyform toothers(e.g.hostingcodeonapublicGitHubrepository)isaviolationoftheacademicintegritypolicy.

• zerotolerance à alloccurrenceswillbereported• Finthecourse• referredtotheSchoolofEngineeringDisciplineCommittee• thiscanleadtoexpulsionfromtheUniversity,aswellaspossibledeportationforinternationalstudents.

• Ifyoucopyfromanyoneintheclassbothpartieswillbepenalized,regardlessofwhichdirectiontheinformationflowed.

6Thisisyouronlywarning.

Page 7: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

COURSEOBJECTIVE• derive,• understand,• implement,• analyze,and• apply(advanced)machinelearningmethods

7

Page 8: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

THISCOURSE

8

MLtechniques

• structuralriskminimization

• MLEvsMAP• unsupervisedlearning• dimensionality

reduction• semi-supervised

learning• graph-basedML

MLmodels

• kernelmethods• GPs• neuralnetworks• naïveBayes• GMMs• PCA/SVD

...seeRoadmap onthecoursewebpage

PracticalML

• multi-classclassification• featureselection• dimensionalityreduction

Contentsmaybesubjecttochanges!

Page 9: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

BACKGROUND&PREREQUISITES

9

• TheoreticalfoundationsofML• CSE417t• keepthecse417tcoursebookLearningfromDataàwewillusesomeoftheeChapters notcoveredincse417t

• Programmingà Python>=3.4

• ThreepillarsofML- probabilitiesandstatistics- matricesandlinearalgebra- multivariatecalculusandoptimization

CSE517a

Page 10: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

ABOUTTHISCOURSE• Take thiscourseif...

• youalreadyknowaboutthetheoreticalfoundationsofML• youarealreadyveryfamiliarwithsimpleMLmethodssuchaslinear

models,perceptron,decisiontrees,andnearest-neighbormethods• youwanttounderstand advancedMachineLearningmethodsand

techniques• youarecomfortablewithadecentamountofmathematics• youarenotscaredofprogramming(alot!)

• Don’ttakethiscourseif...• youonly wanttoapplyMachineLearningmethods(usingWEKAorscikit-

learntoolboxes)• matrices scareyou• youdon’trememberhowtotakederivatives• youwantaneasyA• youhavenot takenCSE417t

10

Page 11: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

PREVIOUSSTUDENTS’COMMENTS• “[...]Requiresagoodknowledgeinmathandderivatives.”

• “ATONofwork,butmostlyworthitforaveryvaluableskill.”

• “greatcourse,butpreparetoworkyourbuttoff.”

• “Thetopicswereprettycomplicatedanddifficulttounderstandquickly.Iwouldhavepreferredaslightlyslowerpace.”

• “It'smostlyamathclass”

11

Page 12: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

WHATISMACHINELEARNING?Machine Learning

Data

ProgramOutput

Computer

Traditional CS:

Machine Learning:

Data

OutputProgram

Computer

Machine Learning

Data

ProgramOutput

Computer

Traditional CS:

Machine Learning:

Data

OutputProgram

Computer12

DataOutput

Computer

Program

Computer

Data

Output

Machine Learning

Training: Testing:

Machine Learning:

Contentsinthisslidemaybesubjecttocopyright.AdoptedfromKillianWeinberger.Thanks,Killian!.

Page 13: CSE 517A –MACHINE LEARNINGm.neumann/sp2019/cse517/01_CourseO… · • A First Course in Machine Learning, Rogers and Girolami, 2nd ed. (We will use this book for readings, mathematical

DEFINITION

Mitchell1997:AcomputerprogramAissaidtolearn fromexperienceEwithrespecttosomeclassoftasksTandperformancemeasureP,ifitsperformanceattasksinT,asmeasuredbyP,improves withexperienceE.

13