一种基于多支持向量机分类器的并行学习算法

A parallel learning algoriflnn based on multiple support vector machine classifiers

为了在并行计算系统中应用支持向量机，提出一种基于多支持向量机分类器的并行学习算法．分析了w-model算法的不足，并在训练过程中采用循环式反馈更新各支持向量机分类器以避免样本的分布状态对各分类器性能的影响，提高各分类器的训练精度．学习过程以平均分类精度为阈值，对部分分类器重新训练，实现对多分类器学习系统性能的全局优化．在UCI标准测试数据集上进行的实验结果表明，循环式反馈能有效地平衡多分类器学习性能相差过大的问题，算法较w-model具有更高的训练效率和分类效率．

In order to apply support vector machine to parallel computation setting, a parallel leaming algorithm based on multiple support vector machine classifiers is proposed. The deficiencies of w-model algorithm are analyzed, Classifiers are updated by circular feedbacks to avoid the potential impact of training samples＇ distribution to the classifiers and improve training accuracy of classifiers during the training. The mean classification precision is taken as a threshold to select parts of classifiers which need be trained again and the performance optimization of the whole learning system is achieved. The experimental results on the UCI standard test datasets show that the proposed circular feedback strategy is effective to balance the classification performance of multiple classifiers, and the proposed algorithm has higher training and classification efficiencies compared with w-model algorithm.