Distributed Techniques for Frozen and Dynamic Multi-Agent Games

The classical minority game (MG) involves n agents (n being an odd integer) that need to take a binary action at discrete time-steps. Each agent´s goal is to be rewarded at every time-step for taking the minority action with respect to the entire agent population. Otherwise it is punished. Agents cannot explicitly communicate with each other and only have access to the number of agents that took a particular action in the last m time-steps. In the local minority game (LMG) agents aim to be in the minority of their immediate neighbors. We have solved the LMG using the distributed stochastic algorithm (DSA). This leads the agent system to a frozen configuration where agents do not change their states over time. We also present a dynamic version of the LMG and its extension to an application in sensor networks, both of which we model as a random Boolean network (RBN)