Making Allocations Collectively: Iterative Group Decision Making under Uncertainty

A major challenge in the field of Multi-Agent Systems (MAS) is to enable autonomous agents to allocate tasks and resources efficiently. This paper studies an extended approach to a problem we refer to as the Collective Iterative Allocation (CIA) problem. This problem involves a group of agents that progressively refine allocations of teams to tasks. This paper considers the case where the performance of a team is variable and non-deterministic. This requires that each agent is able to maintain and update its probabilistic models using observations of each team's performance. A key result is that each agent needs the capacity to store only two or three observations of a team's performance to find near optimal allocations, and a further increase of this capacity will reduce the number of reallocations significantly.