Seminar Announcement

A Study in Support Vector Machines

Todd Munson
Mathematics and Computer Science Division
Argonne National Labs

Monday, August 19, 2002
Bldg. 921, Room 137
10:00 a.m. - 11:00 a.m.

We discuss interior-point and semismooth methods for solving quadratic programming problems with a small number of linear constraints where the quadratic term consists of a low-rank update to a positive semi-definite matrix. Several formulations of the support vector machine, a technique employed by the machine learning community for supervised learning, fit into this category. A related example is the Huber regression problem which can also be posed as a quadratic program with the desired properties.

Support vector machines can be used, for example, when determining whether a tumor is malignant or benign. An interesting feature of the problems we consider is the volume of data, which can lead to quadratic programs with between 5 and 100 million variables and, if written explicitly, a dense Q matrix. The implementations of the two algorithms use linear algebra specialized for the support vector machine application. For the targeted massive problems, all data is stored out-of-core and we overlap computation and I/O to reduce overhead. Results are reported for several linear support vector machine formulations demonstrating that the algorithms developed are reliable and scalable.

Click here to download presentation.

Hosted by Tammy Kolda (tgkolda@sandia.gov).

This seminar is hosted by the Computational Sciences and Mathematics Research Department at Sandia National Labs in Livermore, CA. For more information on this or other events, visit http://csmr.ca.sandia.gov/news.html. Visitors from outside Sandia require advance arrangements in order to attend. For more information, please contact the CSMR office management assistant Doretha Smith at dahall@sandia.gov or (925) 294-4630.