The formation of cell assemblies: a neural network simulation
Michael S. Landy · Deep Blue (University of Michigan) · 1981
This paper is concerned with the formation of Hebbian cell assemblies in a simple pattern learning task. A neural network model has been defined and a large-scale realization of this model has been programmed which stimulates the process of simple pattern learning. Much work was expended on finding viable functions for the dynamics and learning rule. Self-excitatory cell assemblies representing the various fixations of simple figures were formed, and their characteristics investigated. The simulations failed to form assemblies representing the generalized percept behind the separate fixations. The learning rule also failed to provide predictions for fixation sequence when that sequence was, in fact, predictable. An investigation into possible learning rule modifications was carried out, and this yielded a rule which would solve this latter problem. Several problematic characteristics of the system dynamics were discovered. This led to further ideas concerning changes to the dynamics which might allow for more reverberatory assembly activity and more successful learning.