Mechanisms for Dynamic Setting with Restricted Allocations

Abstract

Dynamic mechanism design is an important area of multiagent systems, and commonly used in resource allocation where the resources are time related or the agents exist dynamically. We focus on a multiagent model within which the agents stay, and the resources arrive and depart. The resources are interpreted as work or jobs and are called tasks. The allocation outcome space has a special restriction that every agent can only work on one resource at a time, because every agent has a finite computational capability in reality.

We propose a dynamic mechanism and analyze its incentive properties; we show that the mechanism is incentive compatible. Empirically, our dynamic mechanism performs well and is able to achieve high economic efficiency, even outperforming standard approaches if the agents are concerned about future tasks. We also introduce a static mechanism under the setting of a restricted outcome space; it is proved that the static mechanism is incentive compatible, and its computational complexity is much less than that of the standard VCG mechanism.