Probabilistic Holons for Efficient Agent-Based Data Mining and Simulation

Multiagent systems (MAS) provide a useful approach to distributed problem solving whereby the growing complexity of robust and scalable systems increases the requirements on efficient structures and organisations. A promising approach to organisation of MAS is the theory of holonic multiagent systems (HMAS), which allows different kinds of recursive agent groupings (holons) whose members temporarily collaborate to solve a certain sub-problem. We present a holonic multiagent system based on a probabilistic agent architecture which is efficient for agent-based distributed data mining and simulation. We illustrate its ability to build different kinds of high-performance holons which are useful for different simulation aspects. The holons are built using cloning and merging of probabilistic networks representing the behaviour and knowledge of the agents. Finally, we show their benefits within the project SimMarket concerning supermarket simulation based on a framework for probabilistic HMAS.