DYNAMIC PROPERTIES OF A BIOLOGICALLY MOTIVATED NEURAL NETWORK MODEL

Abstract

We develop a neural network model based on prominent basic features of biological neural networks. The description keeps a simple but coherent link between the subneuronal, neuronal and network levels. In addition, the variables of the model are endowed with realistic numerical values together with their physical units. This permits to reach quantitative significance for the results. To describe the operation of the neuron, a transfer function is used that is believed to convey more biological significance compared to the usual sigmoid transfer function. It is shown that the dynamic properties of the network, which can vary from stability to chaos, are significantly influenced by the choice of the neuron transfer function. Constraints on the synaptic efficacies, as imposed by Dale’s rule, are also shown to modify the dynamic properties by increasing the stability of the network. A simple neural architecture is presented that leads to a controllable time evolution of the network activities.