Specification and implementation environment for neural networksusing communicating sequential processes

摘要

It is pointed out that from a point of view of parallelnprocessing, neural network models belong to the more general class ofncommunicating sequential processes. Sequential operating neuronsncommunicate through input and output connections with other neuronsnworking in parallel. While such a viewpoint brings nothing novel to thentheory of neural models as such, it can be used to make modelndevelopment, simulations, model alterations, comparisons, and testingneasier. This is because the theory of communicating sequential processesnhas an efficient implementation in the Occam language and transputernprocessors. The basic neural network model specification is given,ntogether with code for model implementation. As an example, Kohonen'snself-organizing map model is used