Single-interval learning by simile within a simulated hebbian neural network

AbstractThis paper describes simulations that illustrate how nerve cells can build more complex concepts by transferring and combining the attributes of other concepts in response to stimulus patterns corresponding to similes. The hypothesis described is startling because information transfer is completed within one step or time interval, even when the concepts involved are situated at different levels of separate concept hierarchies and more than one attribute value must be transferred. Such a rapid mechanism for constructing concepts is considered to be a key component of a more general “single-interval” model of concept manipulation that enables complete sentences to be organized during the time required for neural discharge to move directly across a multilayered network. The single-interval model for learning by simile is contrasted with P. H. Winston's explanation, which involves as many as 12 steps. For the proposed mechanism to be successful in one time interval, knowledge must apparently be encoded in strengthened and weakened connections that correspond to “semantic interference patterns”