Adaptively Weighted Large Margin Classifiers for Sufficient Dimension Reduction

• Activity Number: 413
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Andreas Artemiou*+ and Yufeng Liu
• Companies: Michigan Technological University and The University of North Carolina
• Keywords: Sufficient Dimension Reduction ; Inverse regression ; Support Vector Machine ; Adaptive weights ; Robustness

Abstract

Support Vector Machine(SVM) is a popular large-margin classifier. At the same time, sufficient dimension reduction is a powerful idea in dealing with high dimensional data. Recently Li, Artemiou and Li (2011) introduced Principal Support Vector Machine (PSVM), an algorithm which achieves linear and nonlinear dimension reduction under a unified framework by utilizing inverse regression and SVM ideas. Wu and Liu (2012) proposed adaptively weighted large margin classifiers for robust performance of classification in the presence of outliers. In this presentation we describe how the idea of adaptive classifiers can be used in the Sufficient Dimension Reduction framework to improve the performance of PSVM with outliers.