变量可分离函数的RBF神经网络拟合模型及其VC维

RBFNEURAL NETWORK FITTING MODEL TO VARIABLE- SEPARABLE FUNCTION AND ITS VC DIMENSION

摘要: 提出变量可分离函数的径向基函数网络拟合模型（Fitting Model based Radial Basis Function network to Variable Separable Function，VSRBF）及其学习算法并分析VSRBF的VC维．VSRBF是一个由多个子径向基函数网络组成的分工协作系统，由于把高维模型分解为低维模型，与传统径向基函数网络（Based Radial Basis Function Network，RBF）相比， VSRBF 不仅明显地降低了系统复杂性而且网络的收敛速度更快．证明了VSRBF的VC维低于传统RBF的VC维，实验表明VSRBF在处理高维模型的行为明显优于RBF．

Abstract: A Radial Basis Function network fitting model to Variable-Separable function(VSRBF）is presented with its algorithm and its VC dimension is analyzed. VSRBF is a divide-and-cooperate system which is composed of several sub-RBF networks. Since VSRBF decomposes high dimensional model into low dimensional one, to compare with the conventional RBF, its system complexity is reduced remarkably as well as the faster converging speed. It is concluded that the VC dimension of VSRBF is less than that of the conventional RBF, and the experimental result shows that VSRBF performs advantageously comparing with the conventional RBF in high dimensional model.