摘要： 在研究RBF核函数的几何特性和分析SVM数据依赖性改进方法的基础上，提出了支持向量携带数据冗余信息的论点。冗余信息掩盖了所研究对象的特征，影响SVM的性能。基于黎曼几何的SVM数据依赖性改进方法能够剔除支持向量携带的冗余信息，改进SVM的性能。理论分析和实验研究表明，该方法能够有效提高SVM的分类能力和分类速度。

关键词: 核函数, 冗余信息, 支持向量机, 黎曼几何

Abstract: This paper proposes support vectors including redundant information after analyzing geometrical structure of RBF kernel function and data dependent way for improved Support Vector Machine(SVM). Redundant information confuses the law of a learning problem. It can be excluded with data dependent way based on Riemannian geometry for improved SVM. Experimental results show remarkable improvement on classification ability and classification speed of SVM, supporting this idea.

Key words: kernel function, redundant information, support vector machine, Riemannian geometry

中图分类号: 

• TP391