摘要
传统的支撑向量机(SVM)训练速度非常慢,使用RBF核的序列最小优化(SMO)是有效的SVM改进算法。综合网格法和双线形法的优点,提出分段层近法选择参数惩罚因子C和核参数σ2。同时用来训练二维数据,实验证明,SMO算法与传统的SVM算法都使用该法选定参数,在推广识别率方面为同一水平的情况下,运行速度有很大的提高。
Support vector machincs(SVMs) has a low running rate.Sequential minimal optimization(SMO)is one inefficient SVMs improved method.Combining the advantages of grid search method and two-line search method,this paper uses subsection layer approach way to select penalty parameter C and kernel parameter σ2.Experiments perform on two random datas,showed that SMO has the same generalized recognition rate but are much better on running rate against traditional SVMs,both with the best parameters.
出处
《计算机与数字工程》
2007年第8期60-61,81,共3页
Computer & Digital Engineering
关键词
支持向量机
惩罚因子
RBF核
序列最小优化
参数选择
support vector machincs,penalty parameter,RBF kernel,sequential minimal optimization,parameter selection