Measuring the Specialization of Agents in a Multi-Agent System using Entropy and Complexity

Vivia Nikolaidou and Pericles A. Mitkas


Intelligent Data Analysis, Knowledge Discovery, Multi-Agent Systems Evaluation, Specialization


This work discusses the quantification of specialization in a multi-agent system as an evaluation metric. There is currently no commonly accepted evaluation methodology for multi-agent systems and their complexity emphasizes the need to quantify high-level abstract evaluation metrics. Based on previous work, we measure specialization in two dimensions, horizontal and vertical, using Lopez-Ruiz, Mancini and Calbet complexity, Kolmogorov complexity and Shannon entropy to quantify it. We measured both horizontal and vertical specialization in the Trading Agent Competition agents and found it to be a key factor to their success.

Important Links:

Go Back