Contextual Decision Making: From Predictive to Generative
讲座通知 2026 年 7 月 1 日（星期三）， 下午 14:00 - 15:30 信息管理与工程学院308室 上海市杨浦区武东路100号 主题 Contextual Decision Making: From Predictive to Generative 主讲人 L. Jeff Hong（洪流） University of Minnesota L. Jeff Hong received his bachelor’s degree from Tsinghua University and PhD degree from Northwestern University. He is currently Professor of Industrial and Systems Engineering at the University of Minnesota. He was previously Hongyi Chair Professor at Fudan University (2018-2024), Chair Professor of Management Science at City University of Hong Kong (2014-2018), and Professor, Associate Professor and Assistant Professor at the Hong Kong University of Science and Technology (2004-2014). Prof. Hong’s research interests include stochastic simulation, stochastic optimization, machine learning, and risk management. He has published over 100 papers on leading academic journals and conference proceedings, and have won research awards from the Institute for Operations Research and Management Science (INFORMS), Institute of Industrial and Systems Engineers (IISE), and Operations Research Society of China (ORSC) among others. Prof. Hong is currently the Simulation Department Editor of Naval Research Logistics, and an Associate Editor of Management Science, Operations Research, and ACM Transactions on Modeling and Computer Simulation. He was the Simulation Area Editor of Operations Research (2018-2023) and President of INFORMS Simulation Society (2020-2022). 讲座简介 Contextual decision making seeks to optimize decisions using contextual information that characterizes the operating environment and has become increasingly important in modern applications such as personalized services, dynamic pricing, supply chain management, and healthcare operations, where effective actions must adapt to rapidly changing conditions. A dominant approach in the literature is predictive: one first learns the context-dependent objective function from data and then solves the resulting optimization problem once the context is observed. While intuitive and widely applicable, we show that for a broad class of decision problems with underlying convex structure, predictive approaches face a fundamental limitation: no learning method can simultaneously preserve the convexity of the original problem and avoid the curse of dimensionality in high-dimensional context spaces. Motivated by this barrier, we develop a generative framework that learns the conditional distribution of uncertainty given the context and then optimizes decisions using samples drawn from the learned model. We show that this approach naturally preserves the convexity of the original problem while remaining capable of overcoming the dimensionality challenge in high-dimensional contexts, and is particularly effective when sufficient offline data and computational resources are available to support distributional learning. This is a joint work with Weihuan Huang of Nanjing University and Xinyao Li of University of Minnesota. 微信号｜SUFE_RIIS 扫描二维码 关注我们 RIIS RESEARCH INSTITUTE for INTERDISCIPLINARY SCIENCES
上海财经大学（武东路校区）
