Andrew Lim is a Professor in the Department of Analytics and Operations and the Department of Finance at NUS Business School. He is also affiliated with the Institute for Operations Research and Analytics. His research is in the area of stochastic control, optimization under uncertainty, financial engineering, and robust and data driven decision making. From 2002 — 2014, he was on the faculty of the Department of Industrial Engineering and Operations Research at the University of California (Berkeley). He is a past recipient of the National Science Foundation CAREER Award. He serves as an Associate Editor for Operations Research and Management Science, and was previously on the editorial board of the IEEE Transactions on Automatic Control.
| Name of Speaker | Professor Andrew Lim |
| Schedule | 2 September 2022, Friday at 10:00am
(60 minutes talk + 30 minutes Q&A) |
| Venue | Innovation 4.0 Building, level 1, Seminar Room (next to the level 1 café) |
| Link to Register
(Via Zoom) |
https://nus-sg.zoom.us/meeting/register/tZYld-mhpj4tGNWdd6GD8BGY-whthwWbduyq |
| Title | Mechanisms for Coordinating Systems of Decentralized Agents
|
| Abstract | A typical service/operations system is populated by multiple interacting decentralized agents who make decisions that collectively determine system performance. Decentralized agents are hired because they are domain experts, but the aggregate system is usually not efficient if agents optimize in isolation. Coordination is difficult, however, as it requires a decision maker/mechanism designer/principal who can optimize the aggregate system, which is unrealistic when the system is complex and domain experts are needed to control its many parts. We consider these issues in the setting of a service system modeled by a single server queue where the arrival rate is controlled by multiple agents who dynamically sets prices and earn revenue on each arrival, and the service rate by a different agent who is concerned about minimizing service costs. We show that transfer payments between agents can induce decisions that optimize system efficiency even if every agent misspecifies the impact of the other agents in their models. The optimal transfers, however, depend on the “private” models of each agent and can only be directly computed by a “smart” mechanism designer/principal who can optimize the aggregate system. We propose a mechanism for computing the optimal transfers where decentralized agents iteratively share their valuations of shared resources. We show that this algorithm converges to the optimal transfer function at a geometric rate, and provide natural conditions under which it is optimal for agents to report resource valuations truthfully. This algorithm can be implemented without a “smart” mechanism designer/principal, and decentralized agents are not required to share information about their domain of expertise, or even correctly specify the models, actions or number of other agents when optimizing their decisions.
|
