Dynamic neural unit and function approximation

A dynamic model of a neuron for applications in robotics and control is developed. The proposed model, called the dynamic neural unit (DNU), comprises two distinct operations: the synaptic operation, which determines the optimum feedforward and feedback synaptic weights controlling the dynamics of the neuron; and the somatic operation, which determines the optimum gain of the nonlinear activation function for a given task. The function approximation capability of the network of DNUs is described by considering linear and trigonometric operators. Computer simulation results are presented.<>