Communication efficiency in federated learning in IoT environment
Amutha Prabakar Muniyandi, L. Godlin Atlas, N. Deepa, Mahmoud Ahmad Al‐Khasawneh · 2024
Federated Learning (FL) has emerged as a decentralized machine learning paradigm that enhances privacy and efficiency by enabling edge devices within the Internet of Things (IoT). FL allows models to be collaboratively trained without sharing raw data, thereby preserving data privacy. However, communication overhead remains a significant bottleneck in adopting FL in resource-constrained IoT environments. This chapter delves into the challenges and strategies for improving communication efficiency in FL techniques within IoT ecosystems. It begins by identifying key sources of communication overhead, such as frequent model updates and large data payloads. The chapter then explores state-of-the-art techniques, including model compression, sparsification, quantization, and adaptive communication strategies. Special emphasis is placed on optimizing the trade-off between communication costs and model accuracy while accounting for the heterogeneity of IoT devices and networks. The chapter also highlights recent advancements in edge computing and 5G/6G technologies that complement FL by reducing latency and improving scalability. Practical applications and use cases in smart cities, healthcare, and industrial IoT are discussed to demonstrate the real-world impact of these approaches. Finally, the chapter concludes by summarizing open research directions for achieving sustainable and efficient FL systems in dynamic IoT environments.