Multi-agent meeting scheduling with preferences: efficiency, privacy loss, and solution quality

We consider multi-agent systems with preferences, which are just standard multi-agent systems, except that each agent can set some preferences over its local data. This makes these systems more flexible and realistic, since it is possible to represent possibilities, costs, probabilities, preferences, and penalties. However, it also transforms the search for a feasible solution into the search for an optimal, or good enough, solution, since the use of preferences naturally implies the adoption of an optimization criterion, both for each agent and also for the system as a whole. Thus solution quality becomes an important aspect of such systems. Moreover, each agent may want to keep its information as private as possible. Thus privacy loss is another important aspect of such systems. Finally, an obvious crucial aspect is efficiency, since we would like to find a good solution as quickly as possible. In this paper, we study the relations among these three aspects in the context of a multi-agent meeting scheduling system. We do this by first implementing a multi-agent system that incorporates preferences, and then by running experiments to capture the interesting and useful trade-offs among these three aspects.