Single and multiple compartment models in neural networks
J. Hoekstra · AIP conference proceedings · 1998
In this paper some of the most popular neural networks are classified according to their descent of biological models. It is shown that the activation function of the nodes in these networks stems from either the additive activation model or the shunting activation model. In both models the neuron is seen as a single entity (compartment). The models can be integrated by modeling the voltage regulated membrane ion channels with MOS transistors and splitting the neuron into more compartments. SPICE simulations illustrate this. It is then proposed to extend the class of popular neural networks with nodes that have extensive artificial dendrites. These networks are simulated by multi-compartment modeling.