迅速崛起的机器学习技术——支持向量机  

Rapidly developing machine-studying technology-support vector machine

作　　者：张爱 陆有忠[2] 郑璐石[3] 

机构地区：[1]固原师范专科学校物理系,宁夏固原756300 [2]北京科技大学土木与环境工程学院,北京100083 [3]宁夏大学物理电气信息学院,宁夏银川750021

出　　处：《宁夏工程技术》2004年第2期136-140,共5页Ningxia Engineering Technology

摘　　要：支持向量机(supportvectormachine,简称SVM)是近年来在国外发展起来的一种新型机器学习技术,由于其出色的学习性能,该技术已成为当前国际机器学习界的研究热点.与传统的人工神经网络(artificialneuralnetwork,简称ANN)不同,SVM是基于结构风险最小化(structuralriskminimization,简称SRM)原理,而ANN是基于经验风险最小化(empiricalriskminimization,简称ERM)原理.理论和实验表明,SVM不但结构简单,而且具有较好的泛化能力,尤其是对于小样本问题,成功地克服了ANN学习过程中的“过学习”和可能会陷入局部极小问题.另外,SVM算法是一个凸二次优化问题,能够保证极值解是全局最优解.就SVM理论进行了详细综述,旨在引起广大研究者的重视.Support vector machine is a new kind of machine studying technology which developed from abroad in recent years. Owing to its excellent studying features,the technology has been a hot research issue in the international machine studying field nowadays. Compared to traditional artificial neural network(ANN),SVM was based on structural risk minimization. Both the theory and practice showed that SVM was not only simple in structure,but possessed preferable popularization,especially in little sample which overstudying was solved in the process of ANN and part minimization was successfully overcame. Besides,the method for SVM was a convex quadratic programing problem,which guaranteed that found keys were overall optimum ones. The passage summarized SVM theory and compared and expounded some applications in detail,which aimed to arouse the researcher to pay more attention to it.

关 键 词：经验风险最小化 结构风险最小化 最优超平面 支持向量机 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]