报告人
Ce Li
Boston University
时间
2025年9月3日 星期三
下午 14:00-15:00
地点
科研实验大楼1113
Abstract
Information designers, such as online platforms, often do not know the prior beliefs of their receivers. Meanwhile, the receiver can strategically manipulate the designer's learning process about his prior belief. We design learning algorithms for the designer to learn the receiver’s belief over time through repeated interactions, while being robust to the receiver’s manipulation. Based on the learned prior, the designer can compute approximately optimal signaling schemes using a dynamic robustification procedure that we develop. Our algorithms achieve logarithmic regret with respect to two benchmarks: one is the Bayesian persuasion optimality when the designer knows the receiver’s prior belief and the receiver is myopic, and the other is the designer’s optimality when the receiver optimizes his responses globally. We also show the near optimality of our proposed algorithms by proving lower bounds that nearly match the logarithmic regrets of our algorithms. Our work thus provides a learning foundation for the information design with an unknown belief of a strategic receiver.
Biography
Ce Li is a PhD candidate in the Department of Economics at Boston University. Her research interest is at the interface of microeconomic theory and computer science. On the economic theory side, Ce studies information design and mechanism design. On the computer science side, she studies strategic interactions of agents with learning algorithms. Prior to her doctoral studies, Ce obtained her S.M. in Health Data Science from Harvard University. Ce will be on the academic job market in the 2025-2026 academic year.
