Multicategory Support Vector Machine and Psi-learning

• Activity Number: 116
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Yufeng Liu*+ and Xiaotong Shen and Hani J. Doss
• Companies: The Ohio State University and Ohio State University and Ohio State University
• Keywords: classification ; d.c. programming ; generalization error ; margins

Abstract

Many margin-based classification techniques such as support vector machine (SVM) and psi-learning deliver high performance by directly focusing on estimating the decision boundary, as opposed to estimating the conditional probabilities via regression techniques. As a result, multicategory classification is often treated separately from binary classification; no straightforward generalization is possible. In this talk, I will present a novel multicategory generalization particularly for psi-learning and SVM as a by product, without involving estimation of conditional probabilities, retaining their advantage in the binary case. Computational tools are developed for multicategory psi by utilizing differenced convex (d.c.) programming. Finally, numerical examples will be given to examine the operating characteristics of the proposed methodology and to compare psi-learning with SVM for multicategory problems in terms of generalization.