报告题目：Federated Learning: A Distributed Learning Architecture and Its Challenges
报 告 人：吴迪 博士
报告时间：5月21日（周三）早上10：00
报告地点：临江楼A907-908报告厅
主 持 人：许小龙 教授
报告人简介：
Dr. Di Wu is a Senior Lecturer at the School of Mathematics, Physics, and Computing, the University of Southern Queensland (UniSQ). Prior to that, he was a Researcher Fellow at the Australian Institute for Machine Learning (AIML) and School of Computer Science, University of Adelaide, Adelaide, Australia. He has substantial industry experience in large project management, software development, and large system maintenance experience while working on various projects at China Telecom (Global 500), Shanghai. His research area focuses on applying federated learning, AI security and privacy, and trustworthy AI. He has published papers in high-quality refereed books, conferences, and journals, including top-tier venues such as ICLR, KDD, IJCAI, WWW, TKDE etc. He also serves as an associate editor in NLPJ and a reviewer for many high-quality academic conferences and journals, such as ICLR, ICCV, KDD, ACM MM, CoRL, TMC, TNNLS, PR, etc.
报告简介：
Federated Learning (FL) offers a decentralized paradigm for training machine learning models across distributed devices while preserving data privacy. This talk introduces the foundational concepts of FL and highlights its advantages over traditional centralized learning, especially in scenarios constrained by data sharing and computational resources. Despite its potential, FL faces critical challenges including data heterogeneity, communication efficiency, privacy risks, and learning performance degradation. This seminar presents recent advances and practical solutions addressing these issues, including our research contributions on trustworthy FL, privacy-preserving techniques, backdoor defenses in self-supervised FL, and efficient one-shot federated learning under extreme heterogeneity. Through empirical studies and novel frameworks such as BADFSS, FedInverse, and FedHydra, we demonstrate robust solutions for secure and efficient federated learning across diverse applications. The session concludes with future research directions and open opportunities for interdisciplinary collaboration.
欢迎广大师生踊跃参加！
软件学院
2025年5月19日
