ATIAM ML 2010

From IMTR

Instructor: Arshia Cont & Mathieu Lagrange

This course provides an introduction to pattern recognition and statistical learning for the ATIAM Masters at Ircam during the school year 2010-2011. The goal of the course series is to expose students to problem solving tools using machine learning techniques and intuitions behind each approach.

Topics covered include: Bayesian decision theory; parameter estimation; maximum likelihood; Bayesian parameter estimation; conjugate and non-informative priors; dimensionality and dimensionality reduction; principal component analysis; linear discriminant analysis; density estimation: parametric vs. kernel-based methods; Nearest Neighbor methods; mixture models; expectation-maximization; Sequential Learning; HMMs; Computational Auditory Analysis; Source Separation; and musical applications.

Contents 1 Resources 2 Grading 3 Lectures 4 Group Homeworks

Resources

• Duda, Hart and Stork, Pattern Classification, Wiley Interscience, 2000.
• C. Bishop, Pattern Recognition and Machine Learning, Springer, 2006.
• T. Hastie, R. Tibshirani, J. Friedman, Elements of Statistical learning, Springer, 2001. (Also available online)
• E. Alpaydin, Introduction to Machine Learning, MIT Press, 2004.
• L. Wasserman, All of Statistics: A concise course in statistical inference. Springer Verlag, 2006.

Grading

Grades for the ATIAM Machine Learning course will be distributed evenly between two grade options in ATIAM:

1. Group Grade: A grade based on reports of a TP (Travaux Pratique) which will be held on TBD at ENST, including realization of a Matlab application following given instructions on a musical/machine learning problem.

• 20% of this grade comes from the Group Homework assignments given throughout the course.
• This grade will be distributed within the TSM option and constitues 30% of the overall grade (or 6 over 20).

2. A written exam on TBD, lasting one hour.

• This grade will be distributed evenly with the final STIM grade (50%)

Lectures

Date	Topic	Material	Size
04/11/2010	Introduction	slides	(6Mb)
08/11/2010	Generative vs. Discriminative Learning, Bayesian Decision Theory, Maximum Likelihood Group Homework 1 C. Bishop's slides on Probability Distributions	slides	(14Mb)
15/11/2010	Bayesian Parameter Estimation, Kernel Densities, K-NN, EM Algorithm, Clustering, Sequential Learning	slides	(18Mb)
18/11/2010			()

Group Homeworks

#	Topic	Set	Size
1	Maximum Likelihood on Polynomial Regression	PDF	(~400kb)
2	Bayesian Parameter Estimation on Multinomial/histogram problems	PDF	(~650kb)
3	EM derivation for mixtures of exponential distributions	PDF	(~182kb)
4	HMM State Duration Modeling	PDF	(~295kb)