题目：Privacy-Preserving Robust Learning and Decision-Making with Heavy-Tailed and High-Dimensional Data

主讲人：伊利诺伊大学芝加哥分校 周文心副教授

主持人：西南财经大学统计学院 常晋源教授

时间：2025年6月3日（周二）下午14:00-17:00

    2025年6月10日（周二）下午14:00-17:00    

    2025年6月18日（周三）下午14:00-17:00

    2025年6月25日（周三）下午14:00-17:00

地点：西南财经大学光华校区光华楼1103

报告摘要：

Modern statistical learning and data-driven decision-making increasingly rely on sensitive individual-level and organizational data, creating a fundamental tension between data utility and privacy protection. This lecture series presents recent advances in differentially private robust estimation, regression, and prescriptive analytics, with emphasis on heavy-tailed distributions, high-dimensional features, and nonsmooth decision losses. The first part introduces Gaussian differential privacy and related frameworks, highlighting their statistical interpretation and use in privacy-preserving inference. We study robust mean estimation based on Huber loss, noisy gradient descent, Bahadur representations, Gaussian approximation, and private confidence intervals. A central theme is the trade-off among robustness, statistical accuracy, and privacy, controlled by carefully calibrated robustification and noise parameters. The second part focuses on private linear regression under light- and heavy-tailed errors. Topics include noisy clipped gradient descent for low-dimensional models and noisy iterative hard thresholding for sparse high-dimensional models, with attention to how tail behavior affects convergence rates and privacy costs. The final part considers privacy-preserving feature-based newsvendor problems with unknown demand. We discuss convolution smoothing for nonsmooth losses, f-differential privacy, noisy gradient methods, sparse high-dimensional policy learning, and regret analysis. Applications in healthcare, e-commerce, finance, and retail inventory management illustrate how these methods support reliable and privacy-conscious decisions in data-rich environments.

主讲人简介：

Wen-Xin Zhou is an Associate Professor in the Department of Information and Decision Sciences at the College of Business Administration, University of Illinois at Chicago. From 2017 to 2023, he was a faculty member in the Department of Mathematics at the University of California, San Diego. His research interests include high-dimensional statistics, robust learning for heavy-tailed data, nonparametric statistics, neural networks and deep learning, quantile regression methods and beyond.