Neural Network Adaptive Wavelets for Function Approximation

摘要

Based on the wavelet theory, a new type of Wavelet Neural Network (WNN) is presented. For conventional neural networks (NN), the nonlinear activation function is fixed, such as the sigmodal function, hi this paper, the nonlinear function is a linear combination of wavelets, that can be updated during the networks training process. This new type of WNN is applied to function approximation and it exhibits much higher learning ability compared to the conventional one. Furthermore, BP algorithm and QR decomposition based training method is derived for the proposed network.