The use of artificial neural networks for optimal message routing

The authors investigate the possibility of using an artificial neural network as an adaptive message router to provide optimal or near-optimal routing paths. Two different neural network architectures are implemented and tested against a network simulator to ascertain if their performance is adequate to allow them to be used as routing controllers in a network. A network simulator was written to investigate the performance of the JEB net of Jensen, Eshera, and Barash (1990) and the Hopfield net in routing messages. The results obtained using the enhanced JEB network show that the problem of optimal message routing can successfully by solved through the application of neural networks.<>