A mapping approach for designing neural sub-nets

Several investigators have constructed back-propagation (BP) neural networks by assembling smaller, pre-trained building blocks. This approach leads to faster training and provides a known topology for the network. The authors carry this process down one additional level, by describing methods for mapping given functions to sub-blocks. First, polynomial approximations to the desired function are found. Then the polynomial is mapped to a BP network, using an extension of a constructive proof to universal approximation. Examples are given to illustrate the method.<>