Hopfield neural networks for the scheduling of data flow Petri nets
J.F. Balmat, B.P. Abellard, Robert Maifret · 2002
One of the major problems in parallel architectures conception is the scheduling of the tasks, taking into account the temporal and hardware constraints. Data flow Petri nets (DFPN) are a very powerful tool to model this parallelism. In this paper, we propose a methodology applying the Hopfield neural networks to the DFPN scheduling. In DFPN, the conception of parallel computation algorithms is modelled with a graph which describes the operations set. Thus, the principle is to compute an optimal path in an oriented graph, in order to find the optimal computing time of a program with a limited number of resources. The use of neural networks with feedback connections provides a computing model capable of exploiting fine-grained parallelism to solve a rich class of optimization problems and they can achieve high computation rates by employing a massive number of simple processing elements. We describe the resolution method and show that Hopfield like neural networks are very powerful to compute the scheduling of DFPN.>