最小二乘支持向量机的半监督学习算法  

Semi-supervised learning algorithm with a least square support vector machine

作　　者：张健沛[1] 赵莹 杨静[1] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2008年第10期1088-1092,共5页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(6087303760673131);黑龙江省自然科学基金资助项目(F2005-02)

摘　　要：将最小二乘支持向量机引入到半监督学习中,提出了一种最小二乘支持向量机的半监督学习算法.采用最小二乘支持向量机训练混合样本集,利用最小二乘支持向量机训练速度快、效率高等优点有效地克服了目前一些半监督支持向量机学习算法时间代价大、效率低的缺陷.在训练过程中采用区域标注法,减少达到收敛所需要的迭代次数,并给出了SLS-SVM算法具体的数学描述.在人造数据集及实际数据集上的实验表明,最小二乘支持向量机的半监督学习算法可以有效的减少训练时间,提高训练的速度,从而具有更好的推广能力.The least square support vector machine was applied to semi-supervised learning and resulted in a new learning algorithm, a semi-supervised least square support vector machine （SLS-SVM）. The algorithm trained both labeled and unlabeled examples with SLS-SVM, overcoming the limitation of slow learning and low efficiency in other semi-supervised SVM. The algorithm reduced the iteration number needed to reach convergence by using a region labeling rule. A detailed mathematical description of the SLS-SVM algorithm was presented. Experiments on artificial and real datasets showed that the semi-supervised algorithm on SLS-SVM greatly reduces training time, speeds up the training process, and has better generalization performance.

关 键 词：半监督学习 支持向量机 统计学习理论 最小二乘法 

分 类 号：TP301.5[自动化与计算机技术—计算机系统结构]