虞克明（KemingYu），英国伦敦布鲁内尔大学（Brunel University of London）统计学与数据科学首席教授（ChairProfessor）、布鲁内尔大学数学学科研究影像中心主任，以及统计与数据分析硕士研究生课程主任、英国皇家统计协会会士。
虞克明教授已在Journal of the American Statistical Association、Journal of the Royal Statistical SocietyJournal of Econometrics、 Journal of Business & Economics Statistics、 Electronic Journal of Statistics.Statistica Sinica、Bernoulli、 Journal of Multivariate Analysis、 Journal of Time Series Analysis、
Technometrics等国际学术期刊发表论文160余篇。自2019年美国斯坦福大学首次发布全球前2%顶尖科学家排行榜（World's Top 2% Scientifics）至今，虞克明教授一直在榜多年。
虞克明教授已受邀担任Journal of the American Statistical Association、A&CS、The Royal Statistical Society-A、 The Royal Statistical Society-C、Statistical and Its Interface、Journal of Statistical Theory and PracticeReview等国际期刊的副主编（AssociateEditor），受邀担任欧盟科学基金、英国自然科学基金、英国社会科学基金上会评审专家。
报 告 简 介
Estimation has long been a central topic in Statistics. In recent years, learning has become the dominant approach inAIfor tackling problems involving unstructured data.A critical component in both Statistics and Deep Learning isthe choiceofloss function and performance metrics used fortraining andevaluating models.
Loss functions play avital role in developing effective statistical methods (such as least-squares estimation) and achieving success in dep learning applications, including computer vision, network security, and natural language processing-fields in which various security risks have also emerged. Robust loss functions are essential for handling outliers in Statistics and adversarial examples in Deep Learning. However,most existing robust loss functions fail to address asymmetry, which often arises fromskewed or imbalanced data distributions.
This talk introduces a novel Huberised-type asymmetric los function and its corresponding probability distribution, which is shown to follow a scale-mixture of normals. We then propose a new Bayesian Huberised regularisation method for robust regression. This line of research also holds promise for the development of new Bayesian network models.
