Learning internal representations in the Coulomb energy network

The authors introduce a learning algorithm for the N-dimensional Coulomb network which is applicable to multilayer networks. The central idea is to define a potential energy of a collection of memory sites. Then each memory site is an attractor (or repeller) of other memory sites. With the proper definition of attractive and repulsive potentials between various memory sites, it is possible to minimize the energy of the collection of memories. The authors illustrate this procedure with the Coulomb potential and discuss a supervised learning algorithm using this method. This procedure may be applied to each layer of a multilayer network. The method does not depend on the propagation of error through all layers; thus the system is modular, with each layer trainable independently. Finally, this system is applied to a network which learns the binary mapping for XOR.<>