基于自适应协同聚类的支持向量预选算法  

Preselecting Support Vectors Based on Adaptive Cooperative Clustering

作　　者：李仁兵[1,2] 李艾华[1] 王声才[1] 刘太阳[3] 

机构地区：[1]第二炮兵工程学院502教研室,西安710025 [2]中国空气动力研究与发展中心,绵阳621000 [3]第二炮兵工程学院科研部,西安710025

出　　处：《系统仿真学报》2011年第6期1161-1165,共5页Journal of System Simulation

摘　　要：为提高支持向量机在大规模数据集上的训练效率,提出一种基于自适应协同聚类的支持向量预选算法。该方法通过对两类样本进行自适应协同聚类,寻找少量具有协同关系的类中心对,替代支持向量进行训练,从而减少参训样本数量。其中,中心对数量由算法自动确定。与其他支持向量预选算法的对比实验结果表明,自适应协同聚类算法能够在不影响分类性能的情况下,有效提高训练速度,是一种行之有效的快速支持向量预选算法。To improve the training efficiency of support vector machine（SVM） on large scale datasets,a novel method,based on adaptive cooperative clustering（ACC）,was presented to preselect support vectors（SVs）.A few class center pairs（CCPs） with cooperative relationship were acquired by clustering the two classes of samples cooperatively and synchronously.Then,CCPs were used as support vectors approximately to train SVM and so,samples in training were reduced greatly.In addition,the number of CCPs was confirmed adaptively by ACC.Results of contrast experiments between ACC and other SV-preselecting methods suggest that ACC can increase the training speed effectively without affecting classification ability,and it is a novel fast algorithm for preselecting SVs.

关 键 词：支持向量机 自适应协同聚类 预选取 类中心对 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]