A parallel simulation of fully connected neural networks on a WK-recursive topology

The authors evaluate a fixed-size WK-recursive topology for the distributed parallel simulation of fully connected neural networks. To obtain performance improvements in the design of the parallel algorithm, a systolic strategy was used, in order to guarantee the overlapping between elaboration and communication and the locality of communications, so that a routing is not required. After an introduction of the WK-recursive topology, a performance evaluation was undertaken, and the results are compared with those obtained by using as a target topology a ring of processors and a torus of processors. It is shown how the WK-recursive configuration and the systolic strategy lead to performance enhancement for this class of neural networks