Donald W. Dearholt
Department of Computer Science, NMSU
Abstract
The Pathfinder paradigm utilizes pairwise estimates or measures of proximity
to form a family of networks intended to model aspects of the associations
within human semantic memory. This model supports clustering of similar
concepts (and thus higher levels of abstraction) and minimum-cost paths,
thus providing a well-defined associative structure to the concepts within
a domain. Recently, a method of modeling dynamic phenomena by incrementally
constructing a Pathfinder network based upon counting co-occurring concepts
at each sampling time has been developed, utilizing a canonical scenario.
This procedure can be viewed as computing the cumulative consensus over
a set of adaptive agents, in which each agent has certain responsibilities
for the storage of memories of the co-occurring phenomena. This learning
algorithm transforms sequential phenomena into a Pathfinder network representation,
and thus provides a candidate model for the transition from episodic to
semantic memory in humans.
Keywords
co-occurrence, clustering, dynamic systems, adaptive systems, discrete
models, Pathfinder networks, consensus, agents, episodic memory, semantic
memory, associative graphs, learning algorithm, machine learning