Parallelization of Neural Networks using PVM
Résumé
We use Neural Networks (NN) in order to design control architectures for autonomous mobile robots. With PVM, it is possible to spawn different parts of a NN on different workstations. Specific message passing functions using PVM are included into the NN architecture. A graphical interface helps the user spawning the NN architecture and monitors the messages exchanged between the different subparts of the NN. The message passing mechanism is efficient for real time applications. We show an example of image processing used for robot control. Current version used is PVM 3.4.3 for Solaris and Linux