Parallel simulations of Hopfield neural network on distributed-memory multiprocessors

The authors investigate the parallel simulation of the Hopfield model on a digital computer with a multiprocessor. The synchronous and parallel mode of simulation may result in the oscillation of the network, so they suggest a serial update and parallel evaluation policy, which results in the speedup being proportional to the number of processors used while the convergence of the network is guaranteed. A mapping to a message passing multiple-instruction/multiple-data (MIMD) multiprocessor to reduce the computation time is proposed, and two updating sequences of neurons in the multiprocessor are compared and analyzed.<
>