MACHINE LEARNING

People	Lecturer: Aude Billard, Assistant: Basilio Noris
Exam	15% of the Grade will be based on a recitation by the student (see Practicals below) and participation during recitation. The recitation will take place during class. The remaining 85% of the grade will be based on a 40 minutes oral exam (20 minutes preparation, 20 minutes oral defense). The oral will examine the student on the material viewed during the course.
Prerequisites	Strong knowledge in Linear Algebra, Probability and Statistics This course is intended as an advanced class in ML for PhD students and hence focuses on critical reading of the state of the art in ML. MSc student who want to take this course are expected to have taken the introductory ML course given at the master during the winter semester.
Lecture Notes	The lecture Notes are available here . Chapter 5 on kernel Methods is now complete. New figures will be added by end of March. Hardcopies of the lecture notes will be available by middle of April through the librairie polytechnique.
Practicals Software	The course will have 2 hours of practicals centred on the use of the MLDemos software (available at http://mldemos.epfl.ch ). This open-source software allows to visualize and test most of the major algorithms seen in class.
Time and Location	The place and time of the lectures and practical sessions has not been determined yet.

Objective

The aim of machine learning is to construct systems able to learn to solve tasks given a set of examples of those tasks and some prior knowledge about them. Several fundamental strategies have been proposed over the years to cope with this generic setting. These include the "statistical learning theory" and the "Bayesian approach" to machine learning. The goal of this course is to present in a unified way these strategies, their differences and commonalities, as well as the various concepts underlying them. Illustrating these concepts, several machine learning algorithms will also be presented for well-known models such as artificial neural networks, kernel machines, and graphical models.

Practicals and Recitations

The practicals will start with 4 individual practicals on different topics (Kernel Methods, Classification, Regression and Reinforcement Learning). After these, the students will have a choice between a mini-project or a literature survey. The mini-project will entail implementing an algorithm of choice and one or more variants, evaluating the algorithm performance and sensibility to parameter choices. The literature survey will be done on a topic chosen among a list provided in class. The student will read and analyze a choice of recent articles on the topic and provide a summarization of its content. The mini-project or survey will be presented during class in the second part of the semester, and will be the subject of a discussion among the professor and students.

Lecture Schedule and Timetable

Date	Related Chapter from Lecture Notes	Course Slides	Exercises/TP Template	Solutions
23 Feb 2011	Chapter 1 + Annexes	Concepts (classification, regression, density estimation, etc) + Methodology (training/testing sets, leave one out, crossvalidation, ROC, how to measure significance, etc)	Exercises
2 March	Chapter 3	Trees and ensemble methods Decision Trees	Practicals in room TBA
9 March	Chapter 3	Trees and ensemble methods Boosting, Bagging	Practicals in room TBA	--
16 March	Section 2.1-2.2 and Sections 5.1-5.3.	Structure Discovery, Dim. Reduction PCA, Probabilistic PCA, kernel PCA ( slides, )	Exercises , Matlab code for kPCA	Exercises
23 March	Section 2.3, 5.4, 5.5	Structure Discovery, Dim. Reduction Kernel CCA, Kernel ICA ( Slides)	Recitation Matlab Code kCCA , Matlab Code ICA , Matlab Code kICA
30 March	Section 5.6	Support Vector Machine, RVM, SVR (Slides) Complementary Notes on nu-SVM and SVR	Recitation Matlab Code SVM/SVR
6 April	Chapter 4 and Section 5.7-5.8	Bayesian / statistical learning methods Gaussian Processes, Gaussian Mixture Models Complementary Notes on GP - GMM	Recitation
13 April	Chapter 4	Bayesian / statistical learning methods GPLVM, LWPR (Slides) GP Extensions Complementary Notes on GPLVM and LWPR	Recitation
20 April	Chapters 6.3 and 6.4	Connectionist Models and Extensions ANN (perception, convolution NN, feed-forward backprop)	Recitation	--
27 April		Easter Break	--	--
4 May		Connectionist Models and Extensions ANN II	Recitation	--
11 May		Bayesian / statistical learning methods PAC learning	Recitation	--
18 May	Chapter 7.3	Sequential and Time-dependent models RL and gradient methods, POMDP	Recitation
25 May	Chapter 6.10	Sequential and Time-dependent models RNN, Time-Delay NN	Recitation	--
1 June	Chapter 7.2	Sequential and Time-dependent models HMM Class Overview (AB)	Recitation	--

Papers to Read

The list of papers for recitation is now available here .
You can select the paper you must present through the doodle that was sent out the third week of class. You should enter your selection by March 18. Papers will be assigned on a first pick, firt serve basis, based on the doodle pool results.

- "Kernel Methods for Pattern Analysis" by John Shawe-Taylor, Nello Cristianini
   Publisher: Cambridge University Press (June 28, 2004)
   ISBN-10: 0521813972
   ISBN-13: 978-0521813976

- "Pattern Recognition and Machine Learning" by Christopher M. Bishop
   Publisher: Springer; 1 edition (October 1, 2007)
   ISBN-10: 0387310738
   ISBN-13: 978-0387310732

Statistical Learning Methods:

- "Pattern Classification" by Richard O. Duda, Peter E. Hart, David G. Stork
   Publisher: Wiley-Interscience; 2 Sub edition (October 2000)
   ISBN-10: 0471056693
   ISBN-13: 978-0471056690

- "Introduction to Statistical Learning Theory" by Olivier Bousquet ,
   Stephane Boucheron, and Gabor Lugosi
   http://www.kyb.mpg.de/publications/pdfs/pdf2819.pdf

Neural Networks:

- “Spiking Neuron Models” by W. Gerstner and W. M. Kistler,
Publisher:Cambridge University Press, Aug. 2002

- “Hebbian Learning and Negative Feedback Networks” (Advanced Information and Knowledge Processing) by C. Fyfe.
Publisher: Springer.

- Introduction to Neural Networks: A Comprehensive Foundation (2^nd Edition) by S. Haykins.