A parallel neural network simulator

A neural network simulator was written for course-grained parallel processors in order to simulate large neural networks in a fast and inexpensive way compared with available simulators. Neural networks using the back-propagation learning algorithm are simulated on a network of Inmos transputers. The processing distribution, software architecture, and performance results are described. The results show that this approach provides faster learning times and can simulate large networks. Standard optimization techniques were applied to training neural networks and are described in brief. It is shown that a method which updates the weight vector using the optimal learning rate at each training epoch provides shorter training times than the standard method, which uses a fixed learning rate