题目：Resource-Constrained Federated Learning: Fundamental Analysis and Optimizations

主讲：李骏  教授

时间：2021年11月17日9:30

地点：信息楼C308

主办：信息与通信工程学院

专家简介：

李骏，教授，博士生导师，国家特聘青年专家，南京理工大学无线传感网研究所所长，于2015年入选江苏省特聘教授、2018年入选江苏省双创人才、2019年入选江苏省双创团队。致力于通信、计算以及控制融合的分布式人工智能架构与方法及其在工业互联网中的应用相关研究。发表学术论文240余篇，其中IEEE期刊论文130余篇。以第一申请人授权国际专利1项（美国、中国、日本及韩国），并提交国际标准化组织。荣获2013年度中国信息论年会最佳论文奖、2014年度中国信息论年会最佳报告奖、2017年度国际会议EAI 5GWN最佳论文奖、2018年度IEEE Transactions on Communications期刊模范审稿人、2020年度国际会议IEEE ICCT最佳报告奖、并受邀在2020年中国通信大会(ICCC)上做皇冠体育app人工智能安全的特邀报告。现担任IEEE高级会员、IEEE智能制造标准委员会/技术咨询委员会委员、中国电子学会高级会员、中国计算机协会CCF高级会员、南京市多媒体协会副理事长、IEEE Transactions on Wireless Communications期刊副编辑、IEEE Journal of Selected Topics on Signal Processing期刊客座编辑。担任多个国际知名通信会议技术委员会委员，担任国际会议IEEE IOV 2014的宣传主席、IEEE ICIAS 2020的TPC Co-Chair、以及IEEE GlobeCom2020 Workshop的TPC Chair。曾担任IEEE Communications Letters副编辑。

报告主要内容：

With the rapid development of the Internet-of-Things (IoT), data from intelligent devices is exploding at unprecedented scales. Motivated by the advancing computational capacity of wireless end-user equipment (UE), as well as the increasing concerns about sharing private data, a new machine learning paradigm, namely federated learning (FL), has emerged with increasing attentions. FL allows a decoupling of data provision at UEs and model aggregation at a central server, and thus begins to show its potential advantages in many practical areas. In this talk, we focus on FL with resource-constrained settings, performing fundamental analysis on FL convergence, as well as framework design and optimizations. Specifically, the talk mainly consists of three parts. First, from security perspective, we discuss on FL with differential privacy and external model attacks. Then, we introduce a decentralized FL based on blockchain to combat one-point failure issue, followed by performance analysis and solution to model plagiarism problem. In addition, we consider two FL frameworks in wireless scenarios, and improve FL learning performance by jointly optimizing computing and communication resources. As a conclusion of this talk, we will present some possible challenges and future directions in FL.