Local Synaptic Modification Can Lead to Organized Connectivity Patterns in Associative Memory

Abstract

Without their cerebral cortex people seem to be unable to perform the more interesting types of behavior. On the other hand the cortex is anatomically and electrophysiologically surprisingly uniform. So all the different important capabilities that have been attributed: to different cortical areas seem to be achieved by invoking almost the same machinery. How is this possible?

We propose that the cortex is merely a large memory. The organizing principle of this memory is simple and local: local correlations in cortical activity are stored by enhancing the local connectivity between the active elements. This principle (called Hebb’s law) leads to the long-term storage of “preferred” global activity patterns in the cortex (called cell assemblies). Each of these patterns can be activated by any sufficiently large part of it.

Viewed as a retrieval procedure in a memory, this process is known as self-addressing or as autoassociation in the context of associative memories. As a data storage technique, Hebb’s local rule or the corresponding global mechanism of autoassociation turn out to be indeed efficient, even for the purposes of todays large computer memories. As for the cerebral cortex, there is now experimental evidence for variable synaptic connectivities obeying Hebb’s law.