Distributed parameter neural networks via adaptive wavelet

The paper outlines a new neural net (DPNN) for describing brain’s action based on one-dimensional cable theory. While the traditional neural network system only finding its character involving time changing (as HNN, BP etc.), the model-DPNNs (distributed parameter neural networks) are not only the transmitted neurons of time variation, but also the functions of positions by the voltage u(x. I). With the neuroscientific relevance, some bionural features like intermittent conduction and dendritic spike are fixed well by DPNNs which considered as complicated and adaptive devices contract to the functional elementary units. To find the semi-analytical representation of DPNNs, adaptive wavelets are utilized as new microlocalization tools. While maintaining all advantages of wavelet function, the adaptive wavelet offers a viable alternative learning procedure to the orthogonal least squares method (OLS), Adaptive wavelet method develops a fairly general, low-cost multiscale method for neural net optimization.