“Celebrating the 45th Anniversary of the University of Macau”
A Hybrid Sampling-Based and Gradient Descent: Learning Method and Its Applications
Prof. Kairen ZHANG
Associate Professor, Department of Logistics Management Engineering, Southeast University
Date: 14 May 2026 (Thursday)
Time: 10:10-11:10
Venue: E22-G015
Host: Prof. Li XIAO & Prof. Grace Qi FU
Associate Professor in Business Intelligence and Analytics
Abstract
We develop an algorithmic framework for sequential decision-making problems facing unknown uncertainty over multi-periods. In particular, the decision maker (DM) needs to make two decisions each period while learning from the previous realized uncertainty. The framework integrates sample average approximation (SAA) for the first-stage decision with stochastic gradient descent (SGD) for the second-stage decision, and the resulting generic algorithm attains an O(√T) regret bound. We apply and adapt the framework to develop the algorithms for both the repeated settings with applications such as the capacity allocation and final-buy problems where each period the problem resets itself and the dynamic setting with applications such as multi-echelon inventory systems and dual-sourcing problems where state variable(s) couple the periods together. For the latter class of problems, we overcome the technical challenges arising from the impact of decisions propagates through the state variables that affect both the feasible set and cost structures to establish the regret bounds of the developed algorithms. This is achieved by constructing auxiliary functions that help bound the resulting additional component in the overall regret. Managerial Implications: The algorithms developed for various applications demonstrate great performance numerically and outperform several existing ones. When generalized to the case with more than two decisions each period (under some additional assumption), our framework allows flexible allocation of SAA and SGD components, enabling the DM to adaptively manage data and computational efficiency.
Speaker
Prof. Kairen ZHANG is an Associate Professor in the Department of Logistics Management Engineering at Southeast University. He received his Ph.D. degree in systems engineering & engineering management from Department of Systems Engineering & Engineering Management at the Chinese University of Hong Kong in 2016. He also holds a M.S. in management science from Fudan University and a B.S. in mathematics and applied mathematics from Hunan University. His current research interests focus on data-driven inventory control, supply chain management, project management.
All are welcome!
