摘要
基于最小二乘支持向量机变形,得到一个极其简单快速的分类器——直接支持向量机.与最小二乘支持向量机相比,该分类器只需直接求解一个更小规模矩阵的逆,大大减小了计算量,并未降低分类精度.从理论上证明了该矩阵可逆,保证了分类面存在的唯一性.对于线性情形,采用Sherman-Morrison-Woodbury公式降低可逆矩阵的维数,进一步减少了计算复杂度,使其可适用于更大规模的样本集.数值实验表明,新分类器可行并具有上述优势.
Based on the transmutation of least square support vector machine (I.SSVM). an extremely fast and simple classifier direct support vector machine (DSVM) is proposed. Compared with I.SSVM, the classifier only needs to calculate the inversion of a smaller scale matrix, which greatly reduces the computing amount and has no decreasing classification accuracy. The inversion of the matrix is guaranteed theoretically, and hence a unique classifying hyperplane is existed. For linear classification, the Sherman-Morrison-Woodbury formula is employed to reduce the dimension of the matrix and the computing complexity is decreased, so that the classifier can he applied to larger scale dataset. Numerical experiments show that the classifier is easily operated and has the above advantages.
出处
《控制与决策》
EI
CSCD
北大核心
2008年第8期935-937,943,共4页
Control and Decision
基金
国家自然科学基金项目(60574075
60705004)
关键词
支持向量机
分类
线性方程组
最小二乘
Support vector maehine
Classifieation
Linear equation system
Least squares