The Support Vector Machine as a Supervised Classification Technique

Document Type

Presentation Abstract

Presentation Date

11-5-2012

Abstract

Methods of supervised classification seek to segment a data set into classes based on a priori knowledge. Classification techniques are used on a wide range of scientific problems. For example, ecologists classify aerial images in order to determine how landscapes change over time, while the U.S. Post Office uses classification techniques for handwriting recognition. Other, non-imaging examples include spam detection systems for email and patient diagnosis in medicine.

Supervised classification techniques require the user to provide the set of classes in the data set, as well as a training set for each class. The training set consists of a set of measurements for which its class is known. A classifier is built from the training data, which is then applied to new observations.

The support vector machine (SVM) is a well-known method for supervised classification and is well documented throughout the literature. In constructing an SVM classifier, it is possible to formulate it in such a way that a quadratic minimization problem arises, with both equality and bound constraints.

An introduction to classification, with a focus on image data sets, will be presented, followed by a derivation of the SVM classifier that yields the above mentioned optimization problem. Image data sets will be presented to visually demonstrate the SVM classifier and to compare performance of various constrained quadratic solvers, two of which the presenter developed.

Additional Details

Monday, 5 November 2012
3:10 p.m. in Math 103
4:00 p.m. Refreshments in Math Lounge 109

This document is currently not available here.

Share

COinS