A neural network shortest path algorithm for routing in packet-switched communication networks

This paper presents a Hopfield neural network that solves the routing problem in a communication network. It uses mean field annealing to eliminate the constraint terms in the energy function. Since there are no penalty parameters this approach should avoid the problems of scaling. Computer simulations of the neural network algorithm have shown that it can find optimal or near-optimal valid routes for all origin destination pairs in a fourteen node communication network