This course focuses on the problem of classifying patterns into several categories. The theory of supervised learning in artificial neural networks will be compared with statistical pattern recognition and approaches from the theory of generalization. The course is jointly taught by W. Gerstner and M. Seeger.

In addition to a series of interactive computer exercises in JAVA that is available for this course, we also offer a Miniproject.

The authorative course page is in Moodle .

Overview:

I. Introduction.

• The problem of pattern recognition and automatic classification;

II. Artificial Neural Networks

• Simple Perceptrons and Linear Separability;
• Multi-Layer Perceptrons and BackPropagation Algorithm;
• Generalization and Regularization;
• VC dimension: a pedestrian approach
• Applications

III. Optimal decisision boundaries and density estimation

• Classical Statistical Theories - Bayes Rule
• Maximum Likelihood
• EM-algorithm and mixture models

IV. Related Classical and Modern Methods

• Radial Basis Function Networks (RBF)
• Comparison: RBF Networks, Gaussian mixtures, and Fuzzy Logic
• Support Vector Machines

V. Statistical Learning Theory (taught by M. Hasler); see this page

• Informal Introduction
• Definition of the statistical learning problem
• Structural risk minimization
• VC-dimension
• Support Vector Machines and Learning Theory.

TEXTBOOK
R.O. Duda and P.E. Hart and D.G. Stork: Pattern Classification (Wiley)
OR
S. Haykin: Neural Networks (Prentice Hall)
OR
C. Bishop: Neural Networks for Pattern Recognition (Oxford)
AND
V. Vapnik: The Nature of Statistical Learning Theory (Springer)

Back to home page of W. Gerstner
Go to LCN - Brain-Mind Institute - Computer and Communication Sciences - Life Sciences - EPFL