本文构造了一类单隐层神经网络,使其逼近Rd上连续函数的速度达到最佳代数多项式逼近速度,并刻划了该类单隐层神经网络的逼近性质.
In this paper, we construct a class of neural networks with single hidden layer, by which the approximation rate to continuous functions of d variables on any compact subset of Rd is the same as the case by the best algebraic polynomial. Fur- thermore, we characterize the approximation properties of this class of neural networks with single hidden laver.