A Theoretical Neural Integrator

A neural network model is introduced which approximates the behavior of an integrator. It is developed from cells that are physiologically realistic. Requirements for insensitivity of network function to cell threshold distribution are derived, and network performance as the forward-path element of a neuromuscular control system is analyzed. The response of the ``integrator´´ to sustained inputs is derived and shown to approximate the desired proportionality of input magnitude and rate of change of output. Finally, deviations from physiological realism are discussed and the visual accommodation system is described as an example of a biological servomechanism in which a neural integrator seems probable.