管理科学系学术讲座（7月28日）

Packaging waste is a common challenge faced by meal-kit subscription services. Before shipping each box, providers must decide how much and what type of insulated packaging to use. While these materials are essential for preserving product quality, excessive use increases material waste and carbon emissions. As the industry leader, HelloFresh is committed to improving the environmental sustainability of its supply chain. This paper introduces the Contextual Packaging problem and develops an online-learning method that leverages the problem structure to efficiently mitigate overpackaging. Our algorithm, Contextual Frontier-Preserving Elimination (CFPE), takes into account weather, transit conditions, box contents, and other contextual information to adaptively select the packaging configuration for each box. The algorithm achieves theoretically optimal performance, with an O ̃(√T) regret bound unattainable by existing methods, where T is the horizon length. We conduct numerical experiments on real delivery data from HelloFresh and observe that CFPE rapidly identifies contexts in which the current guidelines lead to overpackaging. We propose a methodological advance for the meal-kit industry by enabling providers to learn high-granularity optimal packaging guidelines from heterogeneous delivery data, a necessary step toward reducing overpackaging. CFPE is a scalable data-driven alternative to the current chamber-experiment practice, and can also be used in conjunction with such experiments to refine packaging guidelines. The method provides a mechanism to systematically reduce environmental impact while maintaining service quality, rather than treating sustainability as a constraint external to operations. The regret improvement CFPE achieves over existing algorithms has a direct operational interpretation: fewer exploratory shipments are needed before the algorithm narrows attention to promising packaging actions. CFPE also facilitates future adoption of greener packaging materials by enabling faster recalibration of packaging guidelines. Evelyn Xiao-Yue Gong is BP Junior Faculty Chair and Assistant Professor of Operations Management at the Tepper School of Business at Carnegie Mellon University. Gong’s research develops reinforcement learning and algorithms for supply chains and business operations. Gong completed her Ph.D. in Operations Research at Massachusetts Institute of Technology. Her research has received Honorable Mention for INFORMS Computing Society Harvey J. Greenberg Award, Second Place for INFORMS Service Science Best Paper Award (Socially Responsible Track), Best Dissertation Award at the SCM in the Post-Pandemic and AI Age Conference, CIB AI Fellowship and other honors. Gong has published research works in top-tier journals and conferences including Management Science, Manufacturing & Service Operations Management, and NeurIPS. Gong's commentary has appeared in major media outlets including Wall Street Journal, Bloomberg, Forbes, and Business Insider.