摘要
本文综述了多层前传网络（MLP）及径向基函数网络（RBF）对函数任意精度逼近的能力，比较了两种网络的最佳逼近特性。对激活函数类的扩充作了介绍，并说明有限数值精度对函数逼近能力实现的影响。
The neural network capability of approximation to certain kind of functions and theirderivatives is important. This paper primarily focuses on two types of networks:the multilayer perceptron (MLP) and the radial basis function (RBF) network, and compares theirproperties of universal approximation and best approximation to function. In the later part,. the extension of activation function in the hidden layer is introduced. And with the consideration of numerical limitation,the variation on property of approximation is also proved.
出处
《模糊系统与数学》
CSCD
1998年第4期79-84,共6页
Fuzzy Systems and Mathematics
关键词
MLP网
RBF网
函数逼近
神经网络
MLP
RBF
Approximation
Activation Function
Numerical Accuracy
