Distributed resource allocation in multiagent systems is hard to solve. Since the allocation will be done distributively, agents are not aware of others that use the resources that they need and in what quantity. That is, because the agents do not have access to the entire list of allocations, they can attempt to use resources that are not available. One naive approach is to allow agents to try different allocations repeatedly, so that they can eventually an effective allocation can emerge. However, such a technique is difficult to succeed when the resources are scarce but the number of agents is high. An effective solution to the problem has to allow agents to self-organize intelligently rather than randomly. Accordingly, this paper proposes a communication scheme, where agents are allowed to exchange a small part of their prior knowledge with a few of the agents that they know. We study our proposed approach in relation to existing approaches in the literature and show the positive effects of communication on better resource allocation, especially when the resources are scarce and the agents have a variety of choices for allocation.
展开▼
