摘要
海量数据分类问题是目前SVM学习算法研究的一个热点。传统的SVM方法是通过计算二次规划问题求解的，当训练样本数较大时存在一定的困难。文章介绍一种新的方法，该方法把二分类问题看作特殊的软最小球覆盖问题。提出求解SVM问题的全新的最小球覆盖几何解释，通过软最小球算法进行求解，避免了计算复杂度高的问题。可以用来求解11目标函数海量数据v-SVM分类问题。实验结果表明，文章提出的算法与同类算法相比具有相当的竞争力。
How to solve a large-scale classification is one of the most important problems in SVM （Support Vector Machine） community now. The traditional method for SVM is solved by quadratic programming, which was limited when the number of training sample is too larger. In this paper, we propose a new method which views binary classification problem as a special soft minimum enclosing ball （MEB） problem in computational geometry and can be solved by soft-MEB algorithm. This method can solve v-SVM based on 11 cost function with low computational cost. Some experiments demonstrate the competitive performance of our method with other algorithm.
作者
张敬忠
余频杰
ZHANG Jing-zhong, YU. Ping--fie （PP,. and A1 Lab,Artillery Academy of PLA,Hefei 230031,China）
出处
《电脑知识与技术》
2007年第12期1384-1388,共5页
Computer Knowledge and Technology
