Dynamic Scheduling with Bayesian Learning

Dr. Xiaoshan Peng will provide a seminar hosted by the ISE Department and organized by the CAO at 10:40 am via Zoom on Feb 16.

Title:  Dynamic Scheduling with Bayesian Learning

Abstract: We consider the dynamic scheduling problem of a single-server multiclass queueing system. A class k customer incurs a cost of ck, which is unknown, for each unit of time waiting in the system. The server observes a realization of ck after each service completion of a class k customer and updates her belief about the cost parameter using the Bayes’ rule. The server decides on the optimal scheduling policy that minimizes the expected discounted cost during an infinite horizon. We show that the famous cmu-rule is no longer optimal when the cost parameters are unknown. When the system consists of one class of customers with a known cost parameter and one class with an unknown cost parameter, the optimal scheduling policy follows the Gittins index policy. We provide a closed-form characterization of the Gittins index. We also show that the Gittins index policy is no longer optimal for systems with more than three classes or more than two classes with unknown cost parameters.

Short Bio: Dr. Peng is an assistant professor at the department of Operations and Decision Technologies at the Kelley School of Business, Indiana University. Her research focuses on queueing theory with applications in service and healthcare operations. Her works have been published in Queueing Systems and Operations Research. She obtained her Ph.D. from Booth School of Business at the University of Chicago, an MS in Operations Management from Kellogg School of Management, and a B. Eng. in Industrial Engineering from Tsinghua University. She was runner-up for George Nicholson Student Paper Competition in 2015.