An accelerated sequential minimal optimization method for the least squares support vector machine

Abstract

Least squares support vector machine(LS-SVM) is an important variant of traditional support vector machine, which is used to solve pattern recognition and function prediction. We propose an improved version of the Sequential minimum optimization(SMO) algorithm for training LS-SVM, based on a acclerated grdient method. In this paper we consider adding a new point to capture previous update information. We adopt the idea of Nesterov acceleration method, which gets intermediate points from previous update information and then updates the new iteration point. we show experimentally that the improvement method can significantly reduce the number of iterations, and the training time of LS-SVM can also be reduced in the improvement first-order SMO.