一种改进的序贯最小优化算法

序贯最小优化(SMO)算法是目前解决支持向量机训练问题的一种十分有效的方法,但是当面对大样本数据时,SMO训练速度比较慢。本文分析了SMO迭代过程中目标函数值的变化情况,进而提出以目标函数值的改变量作为算法终止的判定条件。几个著名的数据集的试验结果表明,该方法可以大大缩短SMO的训练时间,特别适用于大样本数据。

An Improved Sequential Minimal Optimization Algorithm

At present sequential minimal optimization (SMO) algorithm is a very efficient method for training support vector machines (SVM). However, the training speed of SMO is very slow for the large-scale datasets. Analyzing the varieties of the objective function in SMO iterations, we propose a novel improved SMO algorithm in this paper, where the changed value of the objective function is taken as the termination condition. Experiments on several benchmark datasets have been done and the results show that the training time of the proposed algorithm is reduced greatly, especially for the large-scale problems.