利用KKT条件与类边界包向量的SVM增量学习算法  

Fast SVM incremental learning algorithm using KKT conditions and between-class convex hull vectors

作　　者：吴崇明[1] 王晓丹[1] 白冬婴[1] 张宏达[1] 

机构地区：[1]空军工程大学导弹学院计算机工程系,陕西三原713800

出　　处：《计算机工程与设计》2010年第8期1792-1794,1798,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(60975026);陕西省自然科学基金项目(2007F19)

摘　　要：为实现对历史训练数据有选择地遗忘,并尽可能少地丢失训练样本集中的有用信息,分析了KKT条件与样本分布间的关系并得出了结论,给出了增量训练中当前训练样本集的构成。为了提高SVM增量训练速度,进一步利用训练样本集的几何结构信息对当前训练样本集进行约减,用约减后的当前训练样本集进行SVM增量训练,从而提出一种利用KKT条件与类边界包向量的快速SVM增量学习算法。实验结果表明,该算法在保持较高分类精度的同时提高了SVM增量学习速度。To utilize the result of the previous training and retain the useful information in the training set effectively, the relationship between the Karush-Kuhn-Tucker （KKT） conditions of support vector machine （SVM） and the distribution of the training samples is analyzed, and the constitution of the current training sample set in the incremental learning is given. To reduce the computational cost of the SVM incremental learning, the current training sample set is reduced from the geometric point of view, and the reduced training sample set is used in the SVM incremental training, therefore an algorithm of fast SVM incremental learning by using the KKT conditions and between-class convex hull vectors is proposed. Experimental results reveal that the given fast SVM incremental learning algorithm has better performance comparing with the conventional SVM incremental learning algorithm.

关 键 词：支持向量机 增量学习 KKT条件 包向量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]