Sphere-Structured Support Vector Machines for Multi-class Pattern Recognition

Abstract

Support vector machines (SVM) are learning algorithms derived from statistical learning theory. The SVM approach was originally developed for binary classification problems. For solving multi-class classification problem, there are some methods such as one-against-rest, one-against-one, all-together and so on. But the computing time of all these methods are too long to solve large scale problem. In this paper SVMs architectures for multi-class problems are discussed, in particular we provide a new algorithm called sphere-structured SVMs to solve the multi-class problem. We show the algorithm in detail and analyze its characteristics. Not only the number of convex quadratic programming problems in sphere-structured SVMs is small, but also the number of variables in each programming is least. The computing time of classification is reduced. Otherwise, the characteristics of sphere-structured SVMs make expand data easily.