Polar backpropagation [artificial neural networks]

A polar backpropagation network is developed for solving the polar classification problem. The architecture is a constrained version of general three-layer backpropagation, such that selected weights correspond to physical entities in the problem, namely, the coordinates of the pole. Heuristic knowledge of the problem is implicit in the network constraints. It is emphasized that the network is not as general as backpropagation, but is useful at solving a specific kind of problem which backpropagation can only partially explain. The polar backpropagation network is not suitable to general classification problems, but is useful at solving the polar classification problem because it locates the pole