Topology-aware Federated Learning in Edge Computing

What is it about?

With the latest advancements in 5G/6G applications, there is an increasing demand for faster processing speed and stricter privacy protection for users' data. This motivated many companies to move some operations to the edge, closer to their users, rather than sending all users’ data to a central location. Federated Learning (FL) has emerged as the perfect solution to facilitate this distributed learning model. Our study reviews various network topology structures that can better support FL, discusses their strengths and weaknesses, and outlines the main challenges and areas for future research. This could help make these technologies more efficient and tailored to different network environments and applications.

Featured Image

Photo by JJ Ying on Unsplash

Photo by JJ Ying on Unsplash

Why is it important?

The topology of the edge network can sometimes be largely overlooked. In this survey, we want to show people that the network topology can be seen as a challenge and a solution. As a challenge, specific topologies impose certain restraints like extra layers of communication and network structures. Whereas a solution in edge computing, topologies offer new ways to address different bottlenecks such as communication overhead, over-dependent to the central server, etc. However, most FL methods typically follow a naive star topology, ignoring the heterogeneity and hierarchy of the volatile edge computing architectures and topologies in reality. It is important to remember that naive star topologies do not always provide the best solution. Multiple topology structures exist in current FL works, and each topology brings its unique benefits.

Perspectives