AI for Scalable and Practical Queueing Network Modeling and Control
发布时间：2026年11月13日
浏览次数：85
发布者: Congcong Zhao
主讲人： Prof. Jing Dong (Columbia University)
活动时间： 从 2026-11-17 10:30 到 11:30
场地： Room 9, Quan Zhai, BICMR
Abstract: Queueing network models are powerful mathematical modeling tools for managing congestion across various applications, yet broader adoption remains limited due to challenges in accessibility and scalability. This talk explores how recent advances in AI can enable more accessible and practical applications of this modeling tool. We first present a model-based reinforcement learning framework, called differentiable discrete event simulation, to solve general queueing network control problems. Our approach enables efficient and scalable policy gradient evaluation using auto-differentiation software. We then present a new data-driven modeling framework based on autoregressive sequence models. This framework allows us to utilize the rich operational data to automatically learn and represent system dynamics. Together, these methods demonstrate how AI may help enhance the reach and usability of queueing models, helping address operational challenges across diverse domains.
