Coordination among Strategic Agents with Asymmetric Information

PASSCODE: eecs3316

Multi-agent decision making is an essential task in various fields across communication networks, energy communities, autonomous driving, and more. Agents in these settings observe partial information and take actions to optimize individual objectives. Complexities arise as the system designer must coordinate agents who operate under varying objectives and information, in order to optimize a collective goal. Such coordination is hindered by the strategic interplay, the necessity to infer the unseen, and the intertwinement between strategy and belief formations, making it a formidable challenge.

This dissertation investigates game-theoretic coordination methodologies. It begins with a static resource allocation problem and proposes a mechanism. This mechanism employs specially designed allocation and tax functions to incentivize agents toward maximal social welfare. The mechanism is compatible with distributed communication networks and equipped with a learning algorithm for Nash equilibrium. Advancing to dynamical decision-making systems, the dissertation evaluates a sequential Bayesian learning setting and demonstrates how a combination of mechanism and information designs avoids the herding behavior of agents and enhances social welfare. Further exploring Markov games, it introduces perfect correlated equilibrium as an innovative coordination solution in environments with asymmetric information.

CHAIR: Professor Achilleas Anastasopoulos