Wavelet Neural Networks
Antonis Alexandridis, Achilleas D. Zapranis · 2014
A wavelet network usually has the form of a three-layer network. The lower layer represents the input layer, the middle layer is the hidden layer, and the upper layer is the output layer. This chapter presents and discusses analytically the structure of the wavelet network. It discusses the initialization phase, the training phase, and the stopping conditions. The chapter presents and evaluates four methods for the initialization of the parameters of a wavelet network. Of the four methods, the simplest is the heuristic method. More sophisticated methods, namely, residual-based selection (RBS), selection by orthogonalization (SSO), and backward elimination (BE), can be used for efficient initialization. During the training phase, the parameters of the wavelet network are changed to minimize an error function between the target values and the wavelet output.