Distributed neural networks microcontroller implementation and applications

Abstract

In this paper it is argued that, for any three-layer perceptron, it is always possible to design an equivalent distributed ANN, wherein the neurons are implemented on the nodes of a communication network, and the synapses between them are established in the communication process. In this approach, neurons are seen as processing and communication entities. Since both local and distributed implementations of a specific ANN are perfectly equivalent, they can use the same set of synapse weights, i.e. a distributed ANN can be trained on a local, equivalent software implementation. Two use cases are presented to demonstrate the validity of the idea.