Modeling the neurodynamics of a biological neuron using a feedforward artificial neural network

A small artificial neural network was trained to be a state variable systems model of a MacGregor biological point neuron. The network was trained with data obtained by integrating the state equations of motion describing the neurodynamic behavior of the point neuron. The architecture and backpropagation training methods are described, and time histories of the state variables of the point neuron are presented to demonstrate that the artificial neural network correctly models the complex spiking behavior of the biological neuron model. This modeling is demonstrated for several different types of step stimulus signals input to the neuron, and also for a ramp-type stimulus input. This artificial neural network representation of the biological neuron is composed of simple sigmoidal artificial neurons.<
>