Hierarchical self-organized learning in agent-based modeling of the MAPK signaling pathway

In this paper, we present a self-organized approach to automatically identify and create hierarchies of cooperative agents. Once a group of cooperative agents is found, a higher order agent is created which in turn learns the group behaviour. This way, the number of agents and thus the complexity of the multiagent system will be reduced, as one agent emulates the behaviour of several agents. Our proposed method of creating hierarchies captures the dynamics of a multiagent system by adaptively creating and breaking down hierarchies of agents as the simulation proceeds. Experimental results on two MAPK signaling pathways suggest that the proposed approach is suitable in stable systems while periodic systems still need further investigations.