用于回归估计的最小二乘广义支持向量机  

Least Square Generalized Support Vector Machines for Regression

作　　者：孙宗海[1] 孙优贤[1] 

机构地区：[1]浙江大学工业控制技术国家重点实验室,浙江杭州310027

出　　处：《系统工程理论与实践》2004年第7期94-97,共4页Systems Engineering-Theory & Practice

基　　金：973项目(2002CB312200)

摘　　要：　提出了一种用于回归估计的最小二乘广义支持向量机.这种最小二乘广义支持向量机的核函数同标准的支持向量机相比没有或者只有很少的限制.将这种用于回归估计的最小二乘广义支持向量机表示成标准的二次规划(QP)问题,采用基于矩阵分裂的超松弛法同投影梯度法相结合的算法来解这一QP问题.根据超松弛法的特点,这一算法可以处理大量数据的情形.Least square generalized support vector machines (LS__GSVMs) are applied to regression estimation. LS__GSVMs' kernel functions have no or few limits when they are compared with standard support vector machines (SVMs). We give a presentation of quadratic programming (QP) problem for the LS__GSVMs. In order to solve the QP problem, we apply the combination of the gradient projection and successive overrelaxation (SOR) based on the matrix splitting. That is, we train the LS__GSVMs with above algorithm. Because SOR handles one point at a time, it can process very large datasets that need not reside in memory.

关 键 词：最小二乘广义支持向量机 回归估计 超松弛法 矩阵分裂 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]