SIP: Signal and Information Processing: Modeling, Filtering and Learning

Autumn Semester 2011

Prof. Hans-Andrea Loeliger

Description

The course is an introduction to some basic topics in signal processing, adaptive filters, detection/estimation theory and machine learning. The course is divided into 3 sections: Linear signal representation and approximation, Learning nonlinear functions and Algorithms for structured models. The following is a brief list of contents for the course. Depending on the course progress, some additional material may be added.

Part I: Linear Signal Representation and Approximation

Hilbert spaces, least squares, LMMSE estimation; 
orthogonal filter banks; 
learning linear functions; SVD, PCA

Part II: Learning Nonlinear Functions

neural networks, kernel methods

Part III: Structured Models and Message Passing Algorithms

factor graphs and message passing algorithms,
hidden Markov models, Kalman filter, RLS,
Gibbs sampling, particle filter

Prerequisites

Recommended: Zeitdiskrete und statistische Signalverarbeitung (Loeliger, 5th Semester) or equivalent.

Lecture Notes, Text Books

Lecture notes (English) will be handed out during the course.

Testat Condition

A confirmation of attendance (''Testat'') is required for admission to the exam. The Testat requires 10 (out of 14) points. One such point may be gained each week, either by presence in the tutorial session or by submitting your own solutions to the problems before the tutorial session of the subsequent week.

The conditions may be reduced in special cases on request.

Examination

30 minutes oral examination in English.