ANN parallelization on a token-based simulated parallel system
Alexandra Ioana Cristea, T. Okamoto · 2003
We believe that parallelism is strongly connected with artificial neural networks (ANN), as biological neural networks are known to make good use of massive parallelism. At present, there has been little research in this direction. We have designed and implemented parallel ANNs on different environments. The best implementation possibilities are given, naturally, by massively parallel computers (dedicated or not). Still, even in the UNIX environment, which is based on the token-passing type of simulated parallelism, speed-ups are possible. In this paper, we demonstrate this statement on a very simple example problem, designed to perform a similar task to that of a feedforward ANN.