Upcoming Seminars

Information Design in Queueing Systems: From Fluid Limits to Human Limits

Date: November 9, 2026

Speaker:

Rouba Ibrahim

University College London

More Info:

Abstract

We study the use of delay information in queueing systems from two perspectives: a macro system-level view and a micro behavioral view. Our approach utilizes a diverse methodological toolkit, ranging from heavy-traffic limits to controlled behavioral experiments with human subjects. By bridging these methodologies, we highlight robust insights that emerge across different modeling techniques. Specifically, we show that more information is not always better and that vague or binary signaling can paradoxically enhance performance. Finally, we address the gap between rigorous mathematical modeling and the nuances of human decision-making, offering a way forward to integrate both into a unified, behaviorally-aware framework for operational design.

Speaker's Bio

I am a professor at the School of Management of University College London, where I currently head the Operations & Technology group. I serve on the editorial boards of Management Science and Queueing Systems as an associate editor, and I serve as an Area co-Editor of the Operations and Supply Chains area at Operations Research. My research interests lie in service operations management. I am especially interested in the operational management of queueing systems, from both mathematical and behavioral perspectives.

More Info:

Abstract

TBD

Speaker's Bio

TBD

Date: November 30, 2026

Speaker:

Flore Sentenac

HEC Paris

More Info:

Abstract

TBD

Speaker's Bio

TBD

Date: December 7, 2026

Speaker:

Kuang Xu

Stanford University

More Info:

Abstract

TBD

Speaker's Bio

TBD

Alignment in Large Language Models: Statistical and Game-Theoretic Perspectives

Date: December 14, 2026

Speaker:

Weijie Su

University of Pennsylvania

More Info:

Abstract

Large language models (LLMs) are predominantly aligned with human preferences through reinforcement learning from human feedback (RLHF). In this talk, we explore the theoretical foundations of LLM alignment through the intertwined lenses of statistics and game theory. First, we show how the current formulation of RLHF induces a systematic bias we call preference collapse, and how this can be mitigated by introducing a tailored regularization term into the reward function. Next, we expose a fundamental bottleneck of reward-based alignment, demonstrating that cyclic human preferences cannot be faithfully represented by scalar reward models such as the Bradley-Terry model. More precisely, we establish that such cyclic inconsistencies give rise to a lower bound on the approximation error of any scalar reward fitting. Shifting to a game-theoretic perspective, we focus on Nash learning from human feedback and establish several social choice desiderata for this approach to alignment, including the preservation of preference diversity through the emergence of mixed strategies. Finally, we show that the zero-sum game approach generally cannot perfectly match a target preference distribution as a unique Nash equilibrium.

Speaker's Bio

Weijie Su is a professor in the Wharton Statistics and Data Science Department and, by courtesy, in the Departments of Mathematics and Computer and Information Science at the University of Pennsylvania. He is currently on leave at OpenAI. He serves as a co-director of the Penn Research in Machine Learning (PRiML) Center. Prior to joining Penn, he earned his PhD from Stanford University in 2016 and his bachelor's degree from Peking University in 2011. His research interests include optimization, statistical foundations of generative AI, privacy-preserving data analysis, and high-dimensional statistics. He serves as an associate editor for Operations Research, Journal of Machine Learning Research, Journal of the American Statistical Association, Annals of Applied Statistics, Harvard Data Science Review, and Foundations and Trends in Statistics. He is currently on the ICML 2026 Organizing Committee as the Scientific Integrity Chair and will serve as the Program Chair for ICML 2027.

Date: January 25, 2027

Speaker:

Alexander Stolyar

University of Illinois at Urbana-Champaign

More Info:

Abstract

TBD

Speaker's Bio

TBD

Google Sites

Report abuse