基于超球结构的支持向量机增量学习算法  

Incremental SVM Learning Algorithm Based on Hyper-sphere Structure

作　　者：郭雪松[1] 孙林岩[1] 徐晟[2] 

机构地区：[1]西安交通大学管理学院,陕西西安710049 [2]合肥工业大学管理学院,安徽合肥230009

出　　处：《运筹与管理》2007年第4期45-49,共5页Operations Research and Management Science

基　　金：国家自然科学基金重点项目(70433003);国家863项目(2003AA413033)

摘　　要：本文首先分析了增量学习过程中支持向量与非支持向量的相互转化问题,而后在此基础上提出了基于超球结构的支持向量机增量学习算法。该算法主要利用超球结构,完成对增量学习中训练样本的选取,进而完成分类器的重构。实验表明,该算法比传统支持向量机增量学习算法具有更高的分类精度。The transformation between support vectors and normal vectors during incremental learning is analyzed firstly. And the incremental SVM learning algorithm based on hyper-sphere structure is put forward consequently. The novel approach reconstructs SVM classifier through the selection of training samples in incremental learning with approach based on hyper-sphere structure employed. The results of experiments show that this algorithm works better than traditional SVM incremental algorithms.

关 键 词：机器学习 增量学习算法 超球结构 支持向量机 

分 类 号：G303[文化科学]