Team theory and neural networks for dynamic routing in traffic and communciation networks

The dynamic-routing problem in traffic and communication networks is addressed. Routing nodes must accomplish the following tasks: (i) generate routing decisions on the basis of local information (i.e. the contents of their queues) and possibly of some messages received from neighbouring nodes, and (ii) compute (or adapt) their routing strategies by measuring local variables and exchanging a small number of messages with neighbouring nodes. The first task leads to regard routing nodes as the cooperating decision-makers of a team organization. The second task calls for a computationally distributed algorithm. Such tasks and the impossibility of solving team functional optimization problems under general conditions suggest that each routing node be assigned a set of multilayer feedforward neural networks able to generate routing decisions. The weights of such neural networks are then adjusted by means of a gradient algorithm based on backpropagation.