题　目：Minibatch Stochastic Gradient Descent-Based Data-Driven Decision-Making Framework for Inventory Management

嘉　宾：Jiameng Lyu, Ph.D. Candidate, Tsinghua University

主持人：花成 副教授 上海交通大学安泰经济与管理学院

时　间：2024年12月6日（周五）10:00-11:30

地　点：安泰楼B207室

内容简介：

Stochastic gradient descent (SGD) has proven effective in solving many inventory control problems with demand learning. However, it often faces the pitfall of an infeasible target inventory level that is lower than the current on-hand inventory level. We address the infeasible-target-inventory-level issue from a new technical perspective -- we propose a novel minibatch-SGD-based framework. Our framework is flexible enough to be applied to a wide range of inventory management problems. By devising the optimal batch scheme, our framework achieves the optimal regret bound for both the general convex case and the strongly convex case in these systems.  

Furthermore, we extend the capability of our framework to learn the optimal policy across various parametric policy classes for more complex inventory systems, such as those with positive lead times. A key challenge in these inventory management problems is the bias in feedback and estimators. Our framework can naturally mitigate such bias through a low-switching update manner. By conducting regret analysis of minibatch SGD with limited switching and an asymptotically unbiased gradient estimator, we successfully established the optimal regret.

演讲人简介：

Jiameng Lyu is a PhD candidate at the Yau Mathematical Sciences Center at Tsinghua University, advised by Prof. Yuan Zhou. He earned his B.S. in Statistics from the University of Science and Technology of China in 2021, graduating with the highest honor, Guo Moruo Scholarship. His research interests center around data-driven decision-making including online learning, online optimization, and statistical machine learning, with an emphasis on the applications to inventory management and revenue management.

欢迎广大师生参加！