讲座时间：2025年12月21日（周日）  10:00

地点： 综合楼644会议室

报告题目：Robust Loss Functions for Deep Learning, Bayesian Robust Quantile Regression, Huberised Regularisation and Robust Quantile Random Forest

报告人简介：  

虞克明，英国伦敦布鲁内尔大学统计学与数据科学讲习教授(Chair Professor)、数学学科研究影响中心主任；英国皇家统计学会会士、英国社科基金(ESRC) 评审专家成员、英国自科基金 (EPSRC)评审专家成员 、欧洲科学基金(ESF) 评审专家成员。目前是《Journal of the Royal Statistical Society-C》副主编，也担任过《Journal of the American Statistical Association, A&CS》、《Journal of the Royal Statistical Society-A》等多家国际SCI、SSCI期刊的副主编。目前他主要从事回归分析、非参数统计、机器学习、贝叶斯推断、大数据及非常小的数据分析等方面的理论和方法研究，是贝叶斯分位数回归方法的开拓者，先后在《Journal of American Statistical Association》、《Journal of the Royal Statistical Society: Series B》、《Journal of Econometrics》、《Journal of Business & Economic Statistics》、《Bernoulli》等统计学顶级刊物上发表论文150多篇。

报告摘要： 

In recent years, several research directions in AI and statistics have gained significant attention, particularly those concerned with robustness. These include the development of robust loss functions to handle outliers and adversarial examples in deep learning, with important applications in computer vision, network security, and natural language processing—domains in which new security challenges continue to emerge. Parallel advances in robust quantile regression and robust regularisation methods further highlight the need for principled approaches capable of withstanding data contamination and model misspecification.

This talk introduces a new loss function that provides a unified Bayesian perspective on these research themes. The proposed framework offers a coherent motivation for robust estimation objectives across prediction, learning, and regularisation, and contributes to a deeper understanding of how robustness can be systematically incorporated into quantile regression and modern machine-learning models.