PARALLEL LEARNING STRATEGIES ON THE EMMA-2 MULTIPROCESSOR

This paper presents a development environment for neural models implemented on ELSAG's multiprocessor Emma-2 (registered trademark). The aim of this environment is to test Neural Net capabilities in a wide range of industrial applications in the field of Pattern Recognition. The choice of Emma-2 as target machine allows a meaningful comparison between the novel and the traditional approaches (e.g. for Character Recognition). The Neural Net environment, developed for the Multi-Layer Perceptron with Accelerated Error Backpropagation, is based on distributing a subset of the examples and replicating the network on each processing element. This approach has shown both a remarkable efficiency and promising recognition performance.