基于闭凸包收缩的最大边缘线性分类器

SVM(support vector machines)是一种基于结构风险最小化原理的分类技术.给出实现结构风险最小化原理(最大边缘)的另一种方法.对线性可分情形,提出一种精确意义下的最大边缘算法,并通过闭凸包收缩的概念,将线性不可分的情形转化为线性可分情形.该算法与SVM算法及其Cortes软边缘算法异曲同工,但理论体系简单、严谨,其中的优化问题几何意义清楚、明确.

Maximal Margin Linear Classifier Based on the Contraction of the Closed Convex Hull

The SVM (support vector machines) is a classification technique based on the structural risk minimization principle. In this paper, another method is given to implement the structural risk minimization principle. And an exact maximal margin algorithm is proposed when classification problem is linearly separable. The linearly non-separable problem can be changed to separable linearly by using the proposed concept of the contraction of a closed convex set. The method in this paper has the same function and quality as SVM and Cortes'soft margin algorithm,but its theoretical system is simple and strict,and geometric meaning of its optimization probem is very clear and obvious.