Fast Single-Shot Multiclass Proximal Support Vector Machines and Perceptions

Abstract

Recently Sandor Szedmak and John Shawe-Taylor showed that multiclass support vector machines can be implemented with single class complexity. In this paper we show that computational complexity of their algorithm can be further reduced by modelling the problem as a multiclass proximal support vector machines. The new formulation requires only a linear equation solver. The paper then discusses the multiclass transformation of iterative single data algorithm. This method is faster than the first method. The two algorithm are so much simple that SVM training and testing of huge datasets can be implemented even in a spreadsheet