Transcription
Introduction to Machine LearningCS 590 and STAT 598A, Spring 2010Instructor: S.V. N. Vishwanathan (email: vishy)http://www.stat.purdue.edu/ vishy/introml/introml.htmlJanuary 12, 2010S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning1 / 17
Class DetailsClasses: Tue/Thurs 9:00 am - 10:15 am LWSN B134S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning2 / 17
TextbookIntroduction to Machine LearningAlex Smola and S.V.N. VishwanathanYahoo LabsSanta Clara–and–Departments of Statistics and Computer SciencePurdue University–and–College of Engineering and Computer ScienceAustralian National UniversityS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning3 / 17
Supplementary TextbookS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning4 / 17
Course DescriptionThis is an introductory course in machine learningYou will learn about a number of basic machine learningalgorithms such ask-meansk-nearest neighborsPerceptronnaive BayesEMYou will also some fairly modern topics such asSupport Vector MachinesGaussian ProcessesExponential FamiliesConditional Random FieldsGraphical ModelsStructured PredictionEmphasis throughout the course will be on connections betweenvarious algorithmsS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning5 / 17
Ideal AudienceWell versed with fundamental statistical concepts such asProbabilityRandom VariablesMean and Varianceetc.Comfortable with statistical algorithms such asLinear and Logistic regressionk-means clusteringetc.Good familiarity with a high level programming language such asC, C , or Python. In a pinch Matlab or R will do (but notrecommended).Interested in learning how to efficiently code algorithms for largescale data analysisS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning6 / 17
PrerequisitesRequiredBasic Probability and ApplicationsMA 511: Linear AlgebraProgramming in some high level languageOr equivalent . . .S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning7 / 17
Grading Policy5 Assignments: 10 points eachCourse project: 25 pointsMidterm: 20 pointsClass Participation: 5 pointsOther policies on the course home page. Please review carefully.S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning8 / 17
Office HoursOffice Hours: 2:00 - 3:00pm Tue or by appointment at HAAS 232S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning9 / 17
Frequently Asked Questions IQ: Will I need to do lots of programming?Ans: Yes. This is a very hands on course and will involvecoding different machine learning algorithms. You shouldbudget significant amounts of time for your assignments.S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning10 / 17
Frequently Asked Questions IIQ: Will I need lots of maths to understand your lectures?Ans: I expect familiarity withLinear AlgebraMultivariate CalculusProbability Theoryas pre-requisites. There will be emphasis on rigor evenwhen learning about machine learning algorithms.Q: Can I meet you anytime I want?Ans: I will definitely be around during office hours. You arewelcome to walk in any other time I am in my office, butdo remember that I generally have busy days. To avoiddisappointment it is best to book a slot via email.S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning11 / 17
Frequently Asked Questions IIIQ: Do you reply to emails?Ans: I try to reply to emails as promptly as possible. If you donot hear back from me within 3 - 4 days then please pingme during the class. Your email may have ended up in myjunk mail folder!Q: Can I solve the HW problems collaboratively?Ans: The course policy clearly says:Group discussions are encouraged to further understanddifficult topics. You may consult with other students abouthomework problems, provided that you indicate suchinformation (whom you consulted with, which problem, towhich extent) on your solution sheet. However, you mustrefrain from getting direct answers from others.Any violation will result in zero credit for the assignment.S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning12 / 17
Frequently Asked Questions IVQ: How do I submit my HW?Ans: For problems which do not involve coding, neatly type orwrite the solution and submit in class. I stronglyencourage the use of LaTeX and discourage the use ofMS Word. For solutions which involve coding, submit aprint out in class and send your code via email before theclass.Q: How will you evaluate the project?Ans: First you need to choose a project topic and discuss itwith me. Then you make a proposal which lays out thewhat you will deliver. After the project you will need tosubmit your code and give a short presentation. You willbe evaluated against:What you promised and how much you delivered.Magnitude of your effort vis-a-vis the rest of the classYour presentationS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning13 / 17
Frequently Asked Questions VQ: Will you post notes for all topics?Ans: Yes for almost all topics except standard ones for which Iwill refer you to chapters in a text book or to otherstandard resources.Q: Will you use slides (e.g. powerpoint) for your lectures?Ans: No. I prefer to lecture on the blackboard. Class notes willbe available for download from the course home pageshortly after the class.S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning14 / 17
Topics (Tentative)ReviewDensity EstimationExponential families of distributionsDirected and Undirected Graphical modelsStructured LearningOptimization for Machine LearningKernelsS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning15 / 17
Background SurveyPlease answer as truthfully as possibleCan help me tailor the lecturesTalk to me if you have any concerns or commentsS.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning16 / 17
Thank You!Questions?S.V. N. Vishwanathan (Purdue University)Introduction to Machine Learning17 / 17
Introduction to Machine Learning CS 590 and STAT 598A, Spring 2010 Instructor: S.V:N. Vishwanathan (email: vishy) . C , or Python. In a pinch Matlab or R will do (but not recommended). Interested in learning how to efficiently code algorithms for large . will refer you to chapters in a text book or to other standard resources. Q: Will you .
