Wavelet networks as an alternative to neural networks
I. Ciuca, Andrew Ware · 2002
The paper presents an alternative to the use of feedfoward neural networks as universal approximators. The alternative is based on the wavelet approximation theory of nonlinear functions. An algorithm from the evolutionary computation class is presented for wavelet network learning which as an optional facility, incorporates the capability for removing irrelevant features from input data in classification applications. The results of a dynamic process forecasting application are also presented.