An Evolutionary Approach to Agent-based Pattern Classification

This paper presents an agent-based approach to pattern classification that intensively uses evolutionary techniques to support supervised learning of classifiers. The paper first provides a formalization of the peculiar sort of agents and multi-agent systems that we consider. Then, a brief description of the adopted evolutionary technique is provided together with a justification for its use. Finally, the proposed approach is embedded into a prototypical, real-world application that offers a solid ground for the concrete assessment of classification performances. The preliminary results that we obtained clearly show that the proposed approach realizes a good trade off between accuracy and speed in a fully distributed and decentralized manner.