An agent-based approach to ANN training

Abstract

In this paper a team of agents applied to train a feed-forward artificial neural networks is proposed, implemented and experimentally evaluated. The approach is based on a new variant of the A-Team architecture. Each agent member of the team is executing its own simple training procedure and it is expected that the whole team demonstrates complex collective behavior. The paper includes a description of the proposed approach and presents the results of the experiment involving benchmark datasets. The results show that the approach could be considered as a competitive training tool. © 2006 Springer-Verlag London.