统计与数据科学系系列学术报告之四百七十七期

时间： 2025年7月15日（周二）10:30-11:30

主持人：复旦大学管理学院统计与数据科学系沈娟副教授

地点：史带楼301室

报告人：Dr. Ying Jin University of Pennsylvania

金滢博士宾夕法尼亚大学

题目： Selective conformal prediction for trusted AI-driven decisions

摘要：Conformal prediction offers distribution-free uncertainty quantification for black-box AI models. However, its standard on-average (marginal) guarantees can be insufficient in decision-making processes usually come with a selective nature.

In this talk, I introduce a conformal selection framework that offers selective inference capabilities for conformal prediction. We focus on applications where predictions from black-box models are used to shortlist unlabeled samples whose unobserved outcomes satisfy a desired property. Conformal selection computes p-values for testing each unobserved outcome, and selects unlabeled samples whose p-values are below a threshold determined by the Benjamini–Hochberg procedure, ensuring finite-sample false discovery rate (FDR) control. I will demonstrate applications in (1) drug candidate screening, and (2) filter trustworthy outputs for LLM alignment, as well as extensions to other selective-inference tasks for trustworthy AI-driven decisions.

个人简介：Ying Jin is currently an Assistant Professor in Statistics and Data Science at the Wharton School, University of Pennsylvania. Prior to that, she was a Wojcicki-Troper Postdoctoral Fellow at Harvard Data Science Initiative from 2024 to 2025, working with Professors José Zubizarreta and Marinka Zitnik at Harvard Medical School. She obtained her PhD in Statistics from Stanford University in 2024, advised by Professors Emmanuel Candès and Dominik Rothenhäusler. Her research centers around statistical uncertainty quantification for black-box AI models, generalizability, distributional robustness, causal inference, and their applications in biomedical discovery and human decisions.