ALGORITHMS FOR MAXIMUM MARGIN CLASSIFIERS
Professor Tom Downs, Department of Electrical and Computer Engineering, University of Queensland, Australia

Abstract: For starters we will sample a choice selection of techniques that have been employed for the solution of the quadratic programming problem that arises in support vector machines.

For the main course, an alternative approach, based upon the maximum stability concept (introduced in the physics community) will be described. This approach has led to the development of the Kernel-Adatron algorithm and we will investigate some variants of this technique.

For dessert, we will look at some recent work that employs a geometrical approach to finding the maximum margin hyperplane.

And, as a form of post-prandial entertainment, we will indulge in a little game of "Hunt the Bayes point".

This seminar was held at the Department of Computer Science, Royal Holloway, University of London on 7 June 1999.

back