Federated Learning in Edge Computing: Security and Privacy Challenges

Edge computing (EC) has emerged as a novel computing paradigm located close to data sources. By offloading computing tasks to the network edge, EC significantly reduces data transmission latency and alleviates the dependence on centralized computing resources. However, it struggles to meet the efficiency requirements of large-scale collaborative training in modern machine learning. Federated learning (FL), as a distributed machine learning framework, offers an efficient pathway for collaborative task execution across multiple nodes, relying on a synergistic computing model of “local training on edge devices and parameter aggregation on the server”. Integrating FL into EC scenarios effectively balances data privacy preservation with the maintenance of system intelligence. Nevertheless, the deep integration of these two technologies introduces complex security and privacy challenges, including the dynamic topology of edge nodes, the interference of potential malicious behaviors on model convergence, and indirect privacy leakage stemming from the inherent parameter exchange mechanisms of FL. Against this backdrop, this article provides an in-depth analysis of the technical integration mechanisms and typical application scenarios of EC and FL. Furthermore, it systematically reviews the security and privacy challenges arising from their convergence and outlines potential directions for future research.