基于申威服务器的联邦边缘学习数据队列优化研究

Research on Optimization of Federated Edge Learning Data Queue Based on Sunway Server

摘要:

物联网中的联邦边缘学习系统被认为是一种有前景的技术，可以在保证客户端隐私的同时减少计算负担。在联邦边缘学习系统中，联邦边缘的数据队列是有限的，且联邦边缘与客户端之间的信道是时变的。因此，选择合适数量的客户端上传数据以保持数据队列的稳定，同时最大化学习精确度，是一项挑战。为了解决这一问题，通过结合联邦边缘学习系统和信息瓶颈理论，对如何高效使用数据队列进行了研究，提出了一种基于申威服务器的数据队列优化方法。具体来说，采用Lyapunov优化理论确定最优选择客户端数，在学习精确度和队列稳定性之间实现平衡。进一步使用信息瓶颈理论在保持数据队列大小不变的情况下，最大化数据的压缩和存储效率。通过仿真实验对该方法的性能进行了评估，结果表明，所提出的方法优于已知的基准方法，提升了数据的存储和处理能力，为物联网系统的设计与优化提供了新的思路。

Abstract:

The Federated Edge Learning (FEEL) system in the Internet of Things (IoT) is regarded as a promising technology that can reduce computational burden while ensuring client privacy. In this system, the data queue at the federated edge is limited, and the channel between the federated edge and clients is time-varying. Therefore, selecting an appropriate number of clients to upload data to maintain the stability of the data queue while maximizing learning accuracy is a significant challenge. To address this issue, a study combines the federated edge learning system with information bottleneck theory to investigate efficient utilization of the data queue, proposing a data queue optimization method based on the Sunway server. Specifically, Lyapunov optimization theory is employed to determine the optimal number of selected clients, striking a balance between learning accuracy and queue stability. Furthermore, information bottleneck theory is applied to maximize data compression and storage efficiency while keeping the data queue size unchanged. Simulation experiments are conducted to evaluate the performance of the proposed method, and the results demonstrate that the method outperforms established benchmark approaches, enhancing data storage and processing capabilities and providing new insights for the design and optimization of IoT systems.