Mathematical Analysis of Large Data Sets

Monday, May 1, 2006, 3:40 p.m. PSA 113

Hans Mittelmann

Dept Math & Stats

Support Vector Machines in Machine Learning


Abstract Very large datasets occur in the area of machine learning (ML). The tasks are having a computer "learn" to read handwriting, to understand speech, to recognize faces, to filter spam e-mail etc. Mathematically, these problems lead to huge optimization problems, of an, however not unfavorable type, namely convex quadratic programs (QP).

The talk starts with an introduction to the Support Vector Machine methods for ML and their mathematical characteristics. Three specific such algorithms are described, the SVMlight algorithm, the SVM-QP method, and the Core-SVM approximation method. A comparison on several large datasets is given. Future work will concentrate on the SVM-QP method in collaboration with its author Katya Scheinberg, IBM Watson Research Center.

For further information please contact: Anne Gelb