RESEMBLANCE AND DIFFERENCES BETWEEN WAVELET NETWORKS AND NEURAL NETWORKS
RESEMBLANCE AND DIFFERENCES BETWEEN WAVELET NETWORKS AND NEURAL NETWORKS

Summary/Abstract: In this paper we discuss the resemblance and differences between different type of computational networks such as Neural Networks (NNs) and Wavelet Networks (WNs). Although some theoretical results have shown the potential of using feedforward neural networks as universal approximators, the implementation of such networks suffers from the lack of an efficient constructive method, both for determining the parameters of the neurons and for choosing the network structure. Wavelet networks have been introduced as an alternative to Neural Networks. Essentially, WNs are feedforward NNs that use wavelets as activation function, instead of the classic sigmoidal family. They combine the good localization properties of wavelets with the approximation abilities of neural networks. A type of back-propagation algorithm is also used for learning, but the adjustable parameters to be optimized during the learning stage are both the connection strengths (weights) and the wavelet parameters (position and scale). The purpose of this paper is therefore to compare the main characteristics of NNs and WNs and to highlight the advantages of using the latter ones.

• Details
• Contents