支持向量机的训练算法综述  

A survey on training algorithms for support vector machine

作　　者：王书舟[1] 伞冶[1] 

机构地区：[1]哈尔滨工业大学控制与仿真中心,黑龙江哈尔滨150001

出　　处：《智能系统学报》2008年第6期467-475,共9页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60474069)

摘　　要：支持向量机(SVM)是在统计学习理论基础上发展起来的新方法,其训练算法本质上是一个二次规划的求解问题.首先简要概述了SVM的基本原理,然后对SVM训练算法的国内外研究现状进行综述,重点分析SVM的缩减算法和具有线性收敛性质的算法,对这些算法的性能进行比较,并且对SVM的扩展算法也进行简单介绍.最后对该领域存在的问题和发展趋势进行了展望.Support vector machines （SVMs） use new methods that originated in statistical learning theory. Training of an SVM can be formulated as a quadratic programming problem. The principles of SVM have been summarized briefly in this paper. The latest developments in SVM training algorithms in domestic and overseas research were reviewed, especially reduction algorithms and algorithms with linear convergence properties. The performance of these algorithms was then compared, and a brief introduction to a proposed extension of them was given. Finally some problems and potential directions for future research are discussed.

关 键 词：统计学习理论 支持向量机 训练算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]