Publisher Summary In order to construct models of neural networks in which there is interaction between the neurons, it is necessary to define the model of synaptic transmission. The model proposed by Taylor is reproduced in this chapter. Permanent learning is only possible in networks that contain synapses with this type of characteristic for which the synaptic transmission, defined as the change in output firing rate produced by unit change in input firing rate, is changed by activity. It has recently been shown that the magnitude of the change in transmission required for learning behavior need only be a very small percentage of the initial or “birth” value. A restriction of the magnitude of the synaptic transmission is obviously required if a neuron with thousands of synapses is not to be overdriven, when a large percentage of excitatory synaptic endings are bombarded almost simultaneously at moderate impulse frequencies. Since an overdriven neuron ceases to produce impulses it represents a serious loss of information. If it is accepted that this is to be avoided, so far as possible there must be restrictions on the maximum resultant stimulation that a neuron can receive. The word resultant is important here since stimulation can be excitatory or inhibitory, or positive and negative. Thus, an excitatory stimulus many times the value required to drive a neuron at its maximum frequency could be almost exactly balanced by an inhibitory stimulus. The probability of this state existing by chance is extremely low, however, and it seems more feasible that it could only be maintained by negative feedback circuits. It is known that negative feedback plays an important role in the position control systems of the spinal cord and there is increasing evidence that the feedback principle is also employed throughout the brain. Analogue computers and control systems also make considerable use of the negative feedback principle and it is in these systems that the closest analogues of many neural circuits are to be found.
Read full abstract