文献类型：会议论文

英文题名：Backpressure-Based Federated Learning Model Scheduling in Edge Computing

作者：Zhang, Hengrun[1,2];Fan, Guisheng[1];Lin, Shuai[3]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[2]East China Normal Univ, Shanghai Key Lab Trustworthy Comp, Shanghai 200062, Peoples R China;[3]Shanghai Inst Technol, Sch Econ & Management, Shanghai 201418, Peoples R China

会议论文集：20th International Conference on Collaborative Computing: Networking Applications and Worksharing-COLLABORATECOM-Annual

会议日期：NOV 14-17, 2024

会议地点：Wuzhen, PEOPLES R CHINA

语种：英文

外文关键词：Backpressure scheduling; Federated learning; Secret sharing; Edge computing

摘要：In this paper, we tackle a backpressure scheduling problem in edge computing networks with a privacy constraint, which requires that two specified packets should not be scheduled to arrive at the same node within a certain time interval. Such a scenario can often be encountered when local models of federated learning are further protected based on secret sharing and secure multi-party computation. We show that such a privacy constraint will lead to a time-varying throughput region. Current throughput-optimal backpressure scheduling strategies may suffer from a severe overall throughput degradation given this constraint. In our algorithm, we enlarge the throughput region in each time slot by introducing a new privacy scale. We further prove that our algorithm can still achieve optimal throughput in each time slot. Simulation results show that compared with existing strategies, our algorithm can effectively reduce the number of privacy collisions in each time slot under the same network configuration and arrival process, which achieves a much larger overall throughput region.