Coalition Formation in Sequential Decision-Making under Uncertainty

As real-world applications of coalition formation continuously evolve, the design of new efficient algorithms that maintain a decent and consistent solution over time is required. Specifically, when agents arrive one at a time, a moderator (i.e., an online algorithm) must decide to which coalition the agent should be assigned, if at all. Each agent may be further accompanied with relevant information (e.g., her set of capabilities, her preferences over the previously disclosed agents), based on which the moderator performs its decisions. Multi-agent systems further encompass uncertainties in a variety of forms: the nature of the agents' participation and arrivals may be probabilistic or even unknown. Additionally, their preferences might be not assured and even incomplete or strategic. This research will thus lay the theoretical foundations for studying the interplay between coalition formation and online, uncertain settings, while characterizing the factors which make the moderator's objective susceptible. Our methods will be further tied to practical applications, specifically ones in physical settings (e.g., task allocation in actual robots).