Deep Learning for Data-Driven Operations: From Newsvendors to Non-Crossing Quantile Processes
Prof. Guohao SHEN
Assistant Professor, Department of Applied Mathematics
The Hong Kong Polytechnic University
Hong Kong
Date: 11 March 2026 (Wednesday)
Time: 10:30-12:00
Venue: E22-G015
Host: Prof. Ziwei MEI, Assistant Professor in Business Economics
Abstract
Quantile regression serves as a fundamental tool in operations research, particularly for problems involving risk and uncertainty, such as the newsvendor problem. This presentation explores the theoretical foundations of using Deep Neural Networks (DNNs) for estimating conditional quantiles. Deviating from black-box heuristics, we provide a rigorous analysis of the excess risk bounds, demonstrating how the convergence rate is influenced by network structure, sample size, and the underlying smoothness of the demand function. In addition, we address the challenge of quantile crossing—a pervasive issue in independent quantile estimation. We introduce a ReQU-based deep neural network framework with a specific penalty term to encourage non-crossing properties. A key contribution of this work is the development of new approximation results for ReQU networks, along with the derivation of non-asymptotic risk bounds for the entire quantile regression process. Collectively, these studies offer a theoretically grounded pathway for leveraging the representational power of deep learning in robust operational decision-making.
Speaker
Prof. Guohao SHEN is an Assistant Professor at the Department of Applied Mathematics of The Hong Kong Polytechnic University. He obtained his Ph.D. in statistics from The Chinese University of Hong Kong. His research interests lie in statistical machine learning and nonparametric statistics with a focus on the foundations of deep learning. His work has been published in statistics and machine learning journals and conferences (including AoS, JASA, JRSSB, Biometrika, JMLR, MS, JoE, NeurIPS, ICML and KDD). He is a recipient of the 2024 John Aitchison Prize in Statistics.
All are welcome!
