基于Gauss核函数的快速构造最小超球算法  

Quick algorithm for minimum spheres construction based on Gauss kernel function

作　　者：吴强[1] 贾传荧[1] 李冰梅[2] 

机构地区：[1]大连海事大学航海学院,轮机工程学院辽宁大连116026 [2]大连海事大学轮机工程学院,辽宁大连116026

出　　处：《大连海事大学学报》2007年第3期42-45,50,共5页Journal of Dalian Maritime University

基　　金：辽宁省科学技术计划基金资助项目(2004221010)

摘　　要：为提高基于超球的支持向量机算法中样本数据较多时的训练速度,提出一种构造最小超球的并行融合算法.该算法将全部训练数据集依据特定策略分割成若干个子数据集,分别对各个子数据集进行训练,对所得到的各子数据集的支持向量与融合数据进行训练,构造最小超球.仿真结果表明,并行融合算法在保证分类精度的情况下,能够显著减少训练时间,提高效率,且支持向量的数目较少.同时也验证了该文对Gauss核函数分析的正确性.A parallel-melt algorithm was developed for minimum hyper sphere＇s construction to speed up the training in the support vector machines based on hyper spheres, such as support vector domain description （SVI）D） and sphere structured support vector machines （sphere structured SVMs）. The method divided the original input training dataset of one class into these subsets were trained respectively then. The some small subsets according to the special strategy and final minimum sphere was constructed by training the support vectors of the subsets and melt data. Simulations were performed on one group of artificial data and one group of real data. The results show that the proposed algorithm can reduce the training time obviously while guaranteeing the classification precision and the analysis of the Gauss kernel function is correct.

关 键 词：最小超球 Gauss核函数 支持向量机 并行融合算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]