Neural Network Unit Density: A Critical Biological Parameter

This paper deals with the problem of the implementation of biological properties or characteristics in theoretical formal neurons. Some of these properties are rarely considered, such as the neural network unit density. We developped a computer model implementing neurons with a high biological plausibility, in which it was possible to modify the unit density of the network i.e the length of the axons. The influence of the neural network unit density on dynamic behavior was observed as a nonlinear function. Large nets that don’t display spontaneous cyclic mode become cycling when their unit density decreases and reaches a threshold value. Quasi instantaneous transition from chaotic to stable cyclic activity was found for a critical density value specific to a given network. It is concluded that the process by which a cyclic mode emerges in a neural network is unit density dependent. This phenomenon is reversible and may offer means of studying dynamic transition in quasirandom networks of threshold neurons.