Information transmission in complex neural networks

Abstract

Computational neural networks are inspired in the brain structure and designed to mimic its intelligence artificially. In particular, they can be used as a tool to explore information transport and processing among interconnected units forming layers. In this article we study how the structure of the network (number of layers) and the state of the connections (edge weights) affect to the information flow. A computational model, that allows us to simulate neurons' collective behaviour, has been used over different neural networks configurations, analysing the variation of some structure features. Results show a strong dependence between the number of connections and the response of the network. Also, we have found a relationship among the edge weights distribution and the propagation of the information from the input node to the output layer.