许文盈 教授 (东南大学)

刘歆 研究员

Distributed Optimization over Networks: Algorithm Design and Privacy- Preserving Schemes

11月16日（周一）14:30-15:30，南楼714

How can a network of agents minimize a sum of local objective functions when each agent knows only its own objective and communicates only with its neighbors? In this talk, I will present a unified framework for designing distributed optimization algorithms. The framework makes explicit how agents combine local updates with information from their neighbors, and how these choices influence convergence to a common minimizer. I will then consider distributed online optimization, where agents must act before their current losses are revealed. I will discuss how communication influences regret, which measures cumulative loss relative to a benchmark chosen in hindsight. Finally, I will examine what an observer can infer about an agent’s local objective from the sequence of messages it sends. This leads to the design of privacy-preserving schemes that limit such inference and to the question of how privacy requirements affect convergence and solution accuracy.

报告人简介：许文盈，安徽安庆人，东南大学青年首席教授，博士生导师，入选国家高层次青年人才，德国洪堡学者，数学学院副院长。 2017年获香港城市大学博士学位；随后在新加坡南洋理工大学、德国洪堡大学PIK研究所从事科学研究。在国际重要学术期刊IEEE TAC、IEEE TPAMI、 Automatica、IEEE TSP等发表论文 80 余篇，以第一作者出版英文专著 1部(Springer 出版社)。主持教育部U40，国家自然科学基金3项（面上2项，青年1项），江苏省杰出青年基金，应邀在ICAISC国际会议和全国复杂网络会议作大会报告。牵头获中国仿真学会自然科学一等奖（排1）、中国人工智能学会-吴文俊人工智能青年科技奖、江苏青年五四奖章、江苏省高层次人才培养计划（“333工程”）、江苏省数学成就奖(奖励研究成果富有意义的且年龄不超过四十周岁的青年学者)等。获全国仿真创新应用大赛全国优秀指导教师等，指导学生获第十四届亚洲控制会议最佳学生论文、世界华人数学家大会创意本科论文奖等。任国际权威期刊IEEE T-II、IEEET-ASE、IEEE TSMCS等期刊编委。