Speaker:

Yiding Feng (The Hong Kong University of Science and Technology)

Time:

• 20:00-21:00 Beijing Time
• December 03, 2025 (Wednesday)

Venue:

四号科研楼A区-518

Abstract:

Modern machine learning models–such as large language models-are increasingly accurate at making predictions, and their outputs are often used by downstream decision-makers to guide actions. For these predictions to be truly useful, however, they must not only be accurate but also reliable. A key criterion for reliability is calibration, which ensures that predicted probabilities align with actual outcomes. In this talk, l present an information-design perspective on calibrated prediction. I will introduce a general framework for comparing the informativeness of different machine predictors, and show how this lens helps us understand the limits and possibilities of calibration. I will then discuss how to optimize predictors in environments where decision-makers incentives may be misaligned with those of the predictor.
Based on the joint work with Wei Tang and Liuhan Qian. Some results have been accepted in SODA 2026。

Speaker Bio:

Yiding Feng is an assistant professor at HKUST IEDA. Previously, he worked as a principal researcher atthe University of Chicago Booth School of Business, and postdoctoral researcher at Microsoft Research New England. He received his Ph.D. from the Department of Computer Science, at Northwestern University in 2021. His research focuses on operations research, economics & computation, and theoretical computer science. He was the recipient of the INFORMS Auctions and Market Design (AMD) Michael H. Rothkopf Junior Researcher Paper Prize (second place).