报告题目:Robust Decentralized Vertical Federated Learning based on Adaptive Huber
报告人:练恒 教授 香港城市大学
报告时间:2026年11月17日14:00-15:00
报告地点:腾讯会议 535569762
校内联系人:赵世舜 [email protected]
报告摘要:Federated statistical learning with data distributed on multiple agents has emerged as a promising paradigm for collaborative model training while preserving data privacy. While horizontal learning with sample splitting has been popularly studied, vertical learning where different agents have access to different features for the same individuals, with rigorous statistical and computational guarantees, has attracted less attention. Motivated also by the fact that robustness is critical in modern learning systems, we investigate vertical federated learning for high-dimensional regression using adaptive Huber’s loss in a decentralized system. We propose to use a dual formulation of vertical federated learning which transforms the optimization problem to one similar to horizontal federated learning in form. Using an elastic net penalty, we establish the linear convergence of the dual decentralized ADMM algorithm, which holds under neither distribution assumption of the data nor constraint on the number of nodes. Under the standard iid setting, we also investigated the statistical estimation rate of the estimator. Some numerical examples are used to demonstrate the performance of the proposed algorithm.
报告人简介:练恒,现任香港城市大学数学系教授,于2000年在中国科学技术大学获得数学和计算机学士学位,2007年在美国布朗大学获得计算机硕士,经济学硕士和应用数学博士学位。先后在新加坡南洋理工大学,澳大利亚新南威尔士大学,和香港城市大学工作。在高水平国际期刊上发表学术论文30多篇,包括《Annals of Statistics》《Journal of the Royal Statistical Society,Series B》、《Journal of the American Statistical Association》《Journal of Machine Learning Research》《IEEE Transactions on Pattern Analysis and Machine Intelligence》.研究方向包括高维数据分析,函数数据分析,机器学习等。