The ultra-low latency requirements of 5G/6G applications and privacy constraints call for distributed machine learning systems to be deployed at the edge. With its simple yet effective approach, federated learning (FL) is proved to be a natural solution for massive user-owned devices in edge computing with distributed and private training data. Most vanilla FL algorithms based on FedAvg follow a naive star topology, ignoring the heterogeneity and hierarchy of the volatile edge computing architectures and topologies in reality. In this paper, we conduct a comprehensive survey on the existing work of optimized FL models, frameworks, and algorithms with a focus on their network topologies. After a brief recap of FL and edge computing networks, we introduce various types of edge network topologies, along with the optimizations under the aforementioned network topologies. Lastly, we discuss the remaining challenges and future works for applying FL in topology-specific edge networks.

翻译：5G/6G应用对超低延迟的要求以及隐私约束，促使分布式机器学习系统部署于边缘。联邦学习凭借其简洁有效的方法，被证明是边缘计算中面向海量用户设备、分布式私有训练数据的天然解决方案。大多数基于FedAvg的基础联邦学习算法采用朴素星型拓扑，忽略了现实中易变边缘计算架构和拓扑的异构性与层次性。本文对现有优化的联邦学习模型、框架及算法进行系统性综述，重点关注其网络拓扑。在简要回顾联邦学习与边缘计算网络后，我们介绍各类边缘网络拓扑及其相应优化方案，最后讨论拓扑特定边缘网络中应用联邦学习所面临的现存挑战与未来方向。