基于自适应步长的支持向量机快速训练算法  

Novel fast training algorithm for SVM based on self-adaptive steps

作　　者：姚全珠[1] 田元[1] 王季[2] 张楠[1] 杨增辉[1] 

机构地区：[1]西安理工大学计算机科学与工程学院,西安710048 [2]西北工业大学计算机学院,西安710072

出　　处：《计算机应用研究》2008年第6期1679-1681,1723,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(50279041);陕西省自然科学基金资助项目(2005F07);陕西省教育厅科学技术研究计划资助项目(07JK339)

摘　　要：支持向量机训练问题实质上是求解一个凸二次规划问题。当训练样本数量非常多时,常规训练算法便失去了学习能力。为了解决该问题并提高支持向量机训练速度,分析了支持向量机的本质特征,提出了一种基于自适应步长的支持向量机快速训练算法。在保证不损失训练精度的前提下,使训练速度有较大提高。在UCI标准数据集上进行的实验表明,该算法具有较好的性能,在一定程度上克服了常规支持向量机训练速度较慢的缺点、尤其在大规模训练集的情况下,采用该算法能够较大幅度地减小计算复杂度,提高训练速度。The training method of SVM is to solve the convex quadratic programming.When the amount of training samples is too large,this method will not work.In order to solve this problem and improve the speed of training SVM,this paper analyzed the nature characteristics of SVM and proposed a kind of algorithm for SVM.The speed of classification was much faster than that of conventional SVM in the condition that the correct rate did not decline.The experiments on the UCI database were done with this algorithm.The experimental results show that it has better performance and partly overcomes the flaw of standard SVM,which was slow in the process of classification.This algorithm can remarkably reduce the computation and increase the speed of classification, especially in the case of large number of support vectors.

关 键 词：支持向量机 序贯最小化 机器学习 自适应步长 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]