论文部分内容阅读
分析了径向基插值代理模型的特点,为了提高代理模型的推广能力,引入了结构风险最小化基本原理;指出基函数满足Mercer条件的径向基插值代理模型,本质上也是一类高维分类超平面;文中推导了径向基插值的VC维度与构成径向基插值函数的常数项的导数关系,为此提出了基于结构风险最小化的径向基插值。最后若干函数测试表明基于结构风险最小化的径向基插值提高了代理模型的推广能力。
In order to improve the promotion ability of agent model, the basic principle of structural risk minimization is introduced. The radial basis interpolation agent model which points out that the basis function satisfies Mercer condition is also a kind of high dimensional classification In this paper, the derivative relation between the VC dimension of radial basis interpolation and the constant term of radial basis function is derived. For this reason, radial basis interpolation based on structural risk minimization is proposed. Finally, several functional tests show that the radial basis interpolation based on minimization of structural risk improves the generalization of the proxy model.