G神经网络函数映射能力的构造性证明

CONSTRUCTIONAL METHOD OF FUNCTION APPROXIMATION OF GNN

该文研究了G神经网络的函数映射能力,给出了前馈G神经网络映射任意G型多项式的构造性证明。采用该文的方法映射同一个多项式,所用的神经元数目可少至以往方法的2/(n+1),其中n是G型多项式的次数。

In this paper, the function approximation of Gelenbe Neural Network (GNN) is discussed and it is proved that GNN can approximate any G-type polynomial by using constructional method. Number of units used by this method can be reduced to 2/(n + 1) of that used by previous methods for a same G-type polynomial, where n is the degree of G-type polynomial.