一种基于SVM的最大间隔分类器的研究  

A study of maximal margin classifier based on SVM

作　　者：方科[1] 陈锋[1] 赵子轩[1] 谭现虎[1] 

机构地区：[1]中国科学技术大学自动化系

出　　处：《电子技术（上海）》2008年第4期77-80,共4页Electronic Technology

基　　金：安徽省教育厅重点自然科学基金项目(2006KJ028A);合肥市重点科技计划资助

摘　　要：训练SVM的本质是解决二次规划问题,在实际应用中,如果用于训练的样本很大,标准的二次型优化技术很难应用。最近有学者将S-K算法和核方法相结合,用L_2代价函数来解决SVM问题。这种算法解决了非线性情况和数据不可分的情况。由于它所需的记忆存储和数据量之间呈线性关系,因此这种算法可以用来解决大规模样本集的训练问题。文章对由S-K算法构造最大间隔分类器进行了研究,并用基于S-K算法的核方法构造C-SVM分类器,并取得了令人满意的效果。The key of training SVM is to solve the quadratic programming problem,however in practical applications,if sample sets for training are very large-scale,it is hard to slove with the nore quadratic. The S-K algorithm and the modified kernel technique due to Friess et al.have been recently conbined to solve SVM with cost function.It solve non-linear case using kernel functions and for non-separable data.The requirement in memory storage is linear to data.This property allows the proposed algorithm to be used for large training problems.This paper study maximal margin classifier based on S-K algorithm,and construct C-SVM classifier with kernel S-K algorithm,gain good results.

关 键 词：支持向量机 核方法 S-K算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]