Models of learning and memory

The perceptron and the hologram are two dissimilar devices which have been advanced as neurological models. It :is shown that there are other and perhaps more plausible models which have properties common to both of these devices. The performance of these intermediate models which are termed Associative Nets is described and analysed statistically. The main similarities and differences between perceptron theory and holography can also be demonstrated; it is possible to demonstrate formal links between the translation invariance in certain types of holography and group invariance in perception. theory. Some single proofs of certain theorems in the latter are also given and some other learning procedures are formulated. It is shown that the important difference between these various models resides in the method used to accomplish a modification. If this modification is an analogue of a neurological change a~ a synaptic level, then it should be possible to qualify the relevance of those models by determining what types of synaptic change can take place in different parts of the nervous system. -Although the evidence is far from complete, it suggests that the neocortex is limited to having one type of synaptic change. Finally, each model is discussed in respect of its neurological plausibility