Wavelet neural networks and receptive field partitioning

The use of wavelet functions as basis functions is proposed. Wavelets have many advantages over other basis functions. Orthonormal sets of wavelets can easily be constructed. Thus, network weights can be computed directly and independently. Wavelets can be used to provide a multiresolution approximation of the discriminant functions and offer localization in space and frequency. These properties are put to good advantage by the proposed method, which constructs a sparse wavelet network by including and positioning wavelets from increasing levels of resolution to maximize the classification score.<
>