Applied Machine Learning

APPLIED MACHINE LEARNING

MSc Course

Objectives

Real-world engineering applications must cope with a large dataset of dynamic variables, which cannot be well approximated by classical or deterministic models. This course gives an overview of a number of alternative methods from Machine Learning for the analysis of non-linear, highly noisy and multidimensional data.

Because machine Learning can only be understood through practice, by using the algorithms, the course is accompanied with practicals during which students test a variety of machine learning algorithm with real world data .

The courses uses the MLDEMOS TOOLBOX that entails a large variety of Machine Learning algorithms.

Instructors

Prof. Aude Billard

LASA Laboratory
Swiss Federal Institute of Technology - EPFL
CH-1015 Lausanne, Switzerland

email: aude.billard@epfl.ch

Office Hours: Monday, 14:00 to 16:00, by appointment. (room ME.A3.464)
Tel: +41 (21) 693.54.64
fax: +41 (21) 693.78.50

Assistant

Basilio Noris
email: basilio.noris@epfl.ch

Room: ME.A4.395
Tel: 021 693 78 24

Time and Location

9:15-11:00am: Class takes place in room CO001 for the whole semester EXCEPT DURING THE FORUM, i.e. on October 7 and 14, where it will take place in room CM011

11:15-12am: Alternate between exercises in the same room as class and practicals. Practicals take place in room MXF014

Grades: 50% Practicals (25% written report, 25% oral presentation) + 50% Written Exam

The dates and topics in the timetable below are indicative, and subject to changes.

The report and code of the practicals must be handed in via email to the assistant before the deadline, delays will be penalized. The deadlines for this academic year will be November 4, at 18h00 (written report) and December 16 (oral presentation)

Solutions and grades for the practicals will be posted 14 days after the practicals.

Timetable

--- (Slides ppt)

Date	Course Slides	Exercises/TP Template	Solutions
23/09/11	Concepts + Methodology (Slides)	Brief Recap of Mathematical Bck Needed for the Class	----
30/09/11	METHODS FOR CORRELATION ANALYSIS Principal Component Analysis (PCA)	Exercises The Matrix CookBook	Sol_ex1
07/10/11	METHODS FOR CORRELATION ANALYSIS Independent Component Analysis (ICA) Hyvarinen & Oja Tutorial	Part I: PCA Assignment I , Software
14/10/11	CLUSTERING K-means, GMM, etc (EM - complement)	Part I: ICA Assignment I , (Matlab Code)
21/10/11	CLASSIFICATION Linear classifiers: GMM + Bayes, LDA	Part I: Clustering Assignment I , Software
28/10/11	Assignment I - continued and completion in room MXF014	Assignment I - continued and completion in room MXF014
04/11/11	CLASSIFICATION SVM (Slides)	Exercises Deadline report on assignment 1: November 4 at 18h00 this report counts for 25% of the total grade	Solutions
11/11/11	CLASSIFICATION Multi-Class Classification, Boosting/Bagging (Slides)	Part II: Binary Classification Assignment II , Software
18/11/11	REGRESSION Linear and weighted Regression, Logistic Regression (Slides)	Part II: Multi-Class Classification Assignment II
25/11/11	REGRESSION Support Vector Regression (SVR), Gaussian Mixture Regression (GMR) (Slides)	Part II: Regression Assignment II - Datasets - Groups
02/12/11	(No class!) Practical II: Classification and Regression in room MXF014 Assignment II	Practical continued
09/12/11	MARKOV-BASED METHODS Hidden Markov Models	Exercices	Solutions
16/12/11	Oral Presentations of Practicals Part II this presentation counts for 25% of the total grade	Oral Presentations of Practicals Part II this presentation counts for 25% of the total grade
23/12/11	Overview (Slides)	Exercises/White Exam	---
27 January 2012: 8h15-11h15 Room INM200	EXAM (written, closed books, 3 hours)	---	---

Lecture Notes

The Lecture notes of the class, entitled Machine Learning Techniques are available at the "reprographie/librairie polytechnique". Note that parts of these lecture notes cover also material of the advanced Machine Learning Course given in the spring semester, see website . Although the advanced course is for PhD students, the class is open to MSc students who have followed Part - I of the course. Sole the material in the lecture notes that corresponds to the materia covered through the slides of the MSc is required for the exam of the MSc course.

MachineLearningDemos Software

A suite of algorithms has been implemented for you in the form of a user-friendly program that allows you to play with data and to study how each method performs in different tasks of classification, clustering and regression. A number of examples in the lecture notes have been generated by this software, and are available as examples in the package below (refer to the Lecture Notes to know which examples are available). The software runs on Windows and requires the .Net Framework 3.0 (which should be installed in your machine already, but if it isn't, pick it up here).

Go to MLDemos Page