The goal of this class is to introduce
the student to the language of graphical models in probability and statistical
problems, and to present some applications in Vision and Coding Theory. Time
permitting, we'll explore the link with the Physical Theory of Spin Glasses.
During the last decade, the formalism of graphs has brought new perspectives in
various aspects of learning theory, probability and statistics. A number of
problems have been reformulated in terms of graphical models. This has lead to
new and significantly less computational expensive exact inference algorithms.
At the same time, it has shown the way to more and more satisfactory
approximate ones.
This formalism carries a strong unifying power, and favors the interaction
between different and unrelated topics.
Several techniques originally introduced for a very specific purpose, once
reinterpreted in the language of graphs, found a much broader application in
various fields. Graphical models philosophy roughly consists in breaking down
complicate problems into simple steps, then finding an efficient way to
coordinate these single operations.
The student will develop detailed analysis and Matlab (or other programming
language) simulations of a few simple examples.