An evolutionary model of a neural network.

A very general model of an idealized neural network is proposed, in which the system is governed by an evolution equation. The equation determines the state of the network at time t greater than 0, if its state at time t = 0 and the kernel (in the equation) are known. It is shown that, by making specific assumptions about the kernal and the initial state, the evolution equation describes a distributed memory having the properties of recognition and association. In this case the model of the distributed memory is identical to that of Anderson and Cooper. Further, it is shown that by a different choice of the kernel, the evolution equation goes beyond the earlier model and is able to describe learning as a dynamic process.