Discovering the Structure of a Self-Routing Interconnection Network with a Recurrent Neural Network 1

A Recurrent Neural Network (RNN) is used to learn the structure of an Interconnection Network (IN) by training with examples of routing messages. This is essentially the same as using RNNs to learn the structure of a synchronous sequential machine for which there are many distinct initial states. The method is illustrated by learning a small Self-Routing Interconnection Network (SRIN) with 6 switches by training a second order Single Layer RNN (SLRNN).