Topology-aware Federated Learning in Edge Computing: A Comprehensive Survey

TL;DR

This survey addresses the problem of deploying federated learning at the edge under heterogeneous, hierarchical, and privacy-constrained network environments. It systematically reviews topology-aware FL approaches, categorizing them by star, tree, mesh, and minor/hybrid topologies, and analyzes them through axes such as data partitioning, update protocols, and data distribution. The work highlights key contributions, including a PRISMA-based methodology, a topology-centered taxonomy, and a synthesis of design insights, benchmarks, and future research directions. The findings demonstrate that topology-aware FL can substantially improve communication efficiency, scalability, and privacy in edge settings, guiding both researchers and practitioners toward topology-informed deployments with real-world impact.

Abstract

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 a natural solution for massive user-owned devices in edge computing with distributed and private training data. FL methods based on FedAvg typically follow a naive star topology, ignoring the heterogeneity and hierarchy of the volatile edge computing architectures and topologies in reality. Several other network topologies exist and can address the limitations and bottlenecks of the star topology. This motivates us to survey network topology-related FL solutions. In this paper, we conduct a comprehensive survey of the existing FL works focusing on network topologies. After a brief overview of FL and edge computing networks, we discuss various edge network topologies as well as their advantages and disadvantages. Lastly, we discuss the remaining challenges and future works for applying FL to topology-specific edge networks.

Figures (14)

• Figure 1: The PRISMA flow diagram with reasoning for tree topology.
• Figure 2: Number of papers for each network topology type.
• Figure 3: Right: An example showing the statistical heterogeneity among different types of clients in FL. Depending on how the clients generate data, the statistical distributions and patterns of data on each device can be very different. Left: A demonstration of system heterogeneity. Three different tiers of edge devices have distinct capabilities of computing, connected by links with different bandwidths.
• Figure 4: Application scenarios of federated learning in mobile edge computing.
• Figure 5: Overview of FL topology types.