关于多层感知器的函数逼近能力

ON THE FUNCTION APPROXIMATION ABILITY OF MULTI LAYER PERCEPTRON

给出了一个关于多层感知器（ＭＬＰ）连续函数逼近能力更为普遍的新结论，表明ＭＬＰ的函数逼近能力与隐层数目及隐层神经元特性函数的选取关系不大。

This paper gives a new conclusion on the function approximation ability of multilayer perceptron (MLP), which indicates that the function approximation ability of MLP is mainly determined by the number of the neurons in hidden layers, and has little relation to the activation function of neurons and the number of hidden layers.