Information Properties of a Neuron

Abstract

Two kinds of mathematical neural models, one simular to a sensory neuron and the other to a ganglion cell, are constructed and their properties in view of information theory are studied. Each of the models is essentially a threshold element, the output of which is a train of pulses emitted at the instants of the stored charge inside the neuron reaching the threshold. The interval of the pulses is a random variable, and a method to give its probability density function is derived. It is seen that the distributions of the output pulses of the models are very simular to the spike discharge patterns of actual neurons. For the sensory neuron model the channel capacity is obtained and thus the maximum capability for a neuron to transmit messages is clarified. The model can also be regarded as a sampling device which samples a continuous signal by firing a pulse every time the signal reaches the threshold. A method to recover the original signal from the intervals of the pulses thus emitted is given.