一种新型支持向量机  

A Sort of New SVM

作　　者：杨强[1] 吴中福[1] 余萍[2] 钟将[1] 

机构地区：[1]重庆大学计算机学院,重庆400030 [2]盐城师范学院计算机系,江苏盐城224002

出　　处：《重庆大学学报（自然科学版）》2005年第2期81-84,共4页Journal of Chongqing University

摘　　要：讨论了现有的用于分类的支持向量机 (SVM)所确定的边界在抗干扰方面的局限性。在此基础之上提出了一种新型支持向量机,即基于边界调节的支持向量机,并利用K-T条件得到了这种支持向量机的对偶目标函数。通过对人工数据集和真实数据集的仿真实验表明,相对于L1-SVM而言,基于边界调节的支持向量机具有更少的支持向量和更好的推广性能。The SVM ’s general theorm and shortcoming in resistance desturbance and noise is discussed. The authors find that these shortcoming which is caused by the traditional separating hyperplane. They also present a define, which is called as adjustable separating hyperplane. Basing the adjustable separating hyperplane, a new sort of SVM is set up, and corresponding quadratic programming dual objective function is obtained as well. Simulation results of artificial and real data show that the sort of SVM based on adjustable margin has less number of support vectors and better generating ability than L1-SVM. So, the sort of SVM has some meaning of theory and realism.

关 键 词：支持向量机SVM 最优超平面 核函数 二次规划 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]