OAO-SVMs的训练时间性能分析及算法改进  

Analysis of OAO-SVMs' Training Time and Its Algorithm Improvement

作　　者：张耿[1] 张桂新[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《计算机技术与发展》2007年第7期24-27,共4页Computer Technology and Development

摘　　要：支持向量机(SVM)算法是统计学习理论中最年轻的分支。结构风险最小化原则使其具有良好的学习推广性。但在实际应用中,训练速度慢一直是支持向量机理论几个亟待解决的问题之一,这一点在SVM向多类问题领域推广时表现的尤为明显。文中将从样本分布与类别数量两方面入手,对传统的SVM多分类OAO算法进行训练时间性能上的分析,并引入分层的思想,提出传统OAO-SVMs算法的改进模型H-OAO-SVMs。通过与其他常见多分类SVMs训练时间的比较表明:改进后的H-OAO-SVMs模型具有更优的训练时间性能。SVM （support vector machine） algorithm is the newest branch of statistic learning theory. Because the structural risk minimiza- tion principle makes SVM exhibit good generalization. But in practice, training slowly is one of the questions which are not solved satisfactorily in the field of SVMs. Moreover, the defection is enlarged when expanding SVMs to multi - category classification. Analyses the training time performance of the traditional OAO - SVMs based on swatch distributing and swatch number, presents the hierarchical OAO SVMs model having the better training speed,and compares it to the others＇ multi-class algorithm based on the SVMs.

关 键 词：支持向量机 多分类算法 训练速度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]