摘要
普通的支持向量机算法在对大规模样本进行分类的时候有着较高的时间代价。随着训练样本数量的增多,支持向量机的训练速度问题将会越发明显,并且成为制约其实际应用的瓶颈。针对此问题提出了超椭球面方法,通过去掉噪声点、冗余点,并保留能明确体现样本在空间分布位置特征的样本点,以达到提高支持向量机对大规模样本训练速度的目的。实验表明,超椭球面法在最大限度保证识别正确率的前提下可以大幅加快支持向量机的训练速度。
The traditional SVM algorithm costs much time in large-scale samples classification. With the increase in the number of training samples, the problem of training speed will become more and more obvious,ristricting SVM＇ s promotion and application. Super Ellipsoid-Surface （SES） is proposed in this paper. SES can optimize the training samples by removing noise points or redundant points, reserving the sample points which can reflect the spatial characteristics of samples, so as to improve the efficiency of SVM. Experiments show that SES can improve the training speed markedly.
出处
《北京信息科技大学学报（自然科学版）》
2012年第2期83-87,共5页
Journal of Beijing Information Science and Technology University
基金
煤层气田地面集输信息集成及深度开发技术(011ZX05039-004-02)
