主讲人：朱仲义 复旦大学教授
时间：2026年9月23日14：00
地点：国际交流中心明远厅
举办单位：必威西汉姆联
主讲人介绍：复旦大学统计与数据科学系教授，博士研究生导师；曾任中国概率统计学会第八、九届副理事长，国际著名杂志“Statistica Sinica”副主编； “应用概率统计”“中国科学：数学”杂志编委；现为国际数理统计学会当选会员，现担任“数理统计与管理”杂志编委和国际顶级统计杂志JASA的副主编。专业研究方向为：纵向数据(面板数据)模型；分位数回归模型，机器学习等。主持完成国家自然科学基金面上项目七项、国家社会科学基金一项，作为子项目负责人完成国家自然科学基金重点项目二项，重大项目子项目一项，目前主持国家自然科学基金重点项目一项。近几年发表论文150多篇(其中包括国际四大统计和机器学习顶级刊物等SCI论文80多篇)。2015年获得教育部自然科学二等奖。
内容介绍：Federated learning has attracted significant attention as a distributed framework for training personalised models on multi-source heterogeneous data. However, most existing approaches are unable to handle scenarios where subgroup structures coexist alongside within-group heterogeneity while providing formal differential privacy guarantees. In this paper, we propose a differentially private federated learning algorithm that accommodates heterogeneous tasks by exploiting a latent subgroup structure. Specifically, our method partitions tasks into subgroups to address substantial between-group differences while enabling efficient information sharing among similar tasks within each group. Furthermore, we integrate the Huber loss and a group version of NoisyHT to tackle the challenges of high dimensionality, heavy-tailed distributions, and privacy protection. Theoretically, we establish differential privacy guarantees, derive non-asymptotic error bounds, and provide recovery guarantees for the latent cluster structure. Extensive simulation studies and a real-data application further demonstrate the effectiveness and adaptability of our approach.