演讲人： Jiaye Teng Shanghai University of Finance and Economics
时间： 2025-04-23 14:00-2025-04-23 15:30
地点：FIT 1-222
内容：
In this talk, I will first provide a brief overview of conformal prediction, covering its core algorithms and theoretical guarantees. As a prominent framework in uncertainty quantification, conformal prediction has gained attentions for its simplicity and theoretical foundations. We will begin by discussing the practical deployment of conformal prediction—addressing when and how to apply it in real-world scenarios. I will then delve into its theoretical guarantees, focusing specifically on the coverage properties of the confidence bands it produces. Moving beyond the foundational framework, I will introduce a series of works aimed at improving the precision and reliability of the confidence bands returned by conformal prediction. These include: (a) enhancing precision under neural network regimes; (b) modifying the confidence bands without changing the training process; (c) investigating the evaluation metrics regarding the reliability requirements; and (d) developing variants tailored to different data structures.
个人简介:
Jiaye Teng is currently a Tenure-Track Assistant Professor in the School of Statistics and Data Science at Shanghai University of Finance and Economics. He earned his doctoral degree from the Institute for Interdisciplinary Information Sciences (IIIS), Tsinghua University, in 2024. His research interests span broadly theoretical machine learning, especially generalization theory and conformal prediction. He was honored with Outstanding Graduate at Tsinghua University (2024) and Outstanding Doctoral Dissertation at Tsinghua University (2024), and CCF Incentive Program for Doctoral Dissertations in Theoretical Computer Science (Nomination, 2025). He serves as the organizer of the FAI-Seminar (www.fai-seminar.ac.cn), and his homepage is www.tengjiaye.com.
