Support Vector and Kernel Methods For Pattern Analysis

• Activity Number: 105
• Type: Invited
• Date/Time: Monday, August 4, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Nello Cristianini*+
• Companies: University of California, Davis
• Keywords: kernel methods ; support vector machines ; statistical learning theory ; bioinformatics ; pattern analysis ; pattern recognition

Abstract

Kernel-based learning methods are a class of algorithms for pattern analysis, whose best known element is the Support Vector Machine, based on the idea of replacing inner products by Mercer Kernels. This substitution enables them to operate in high-dimensional spaces, implementing the following simple idea: first embed the data into some vector space, and then detect linear relations in such a space. If the kernel is suitably chosen, points that are "similar" for the task at hand will be mapped to nearby positions, and "different" points will be far apart. The large impact on the field of machine learning is due to many reasons: convex optimization problems in cases where neural networks have been used in the past; intriguing connection with the theory of RKHS; and, because kernels have been recently developed to deal with data as diverse as sequences, graphs, text documents, algorithms have been developed to deal with problems as diverse as classification, ranking, and correlation analysis. Statistical concepts are routinely used in their design, and application range from bioinformatics to text analysis.