Efectos de la comunicacián directa entre agentes en los algoritmos de agrupación de clases basados en el comportamiento de insectos sociales

Abstract

Data clustering algorithms that are based on the behavior of social insects are knowledge discovery tools that rely on stigmergic commmunication, i.e., on the indirect communication through local environment modifications. Inspired by some aspects of trophallaxis, or liquid food exchange among nestmates, we study two direct communication strategies among agents in these kind of algorithms: (i) memory sharing and (ii) environmental maps sharing. The goal of information exchange among agents is to improve the final clustering quality. The effect of the addition of these communication mechanisms is evaluated by comparing the quality of the obtained clustering with direct communication and without communication (besides stigmergic communication). It is also shown that the results depend on the environment agent density and the exchanged information utility. © AEPIA.