Energy-efficient federated learning algorithms
Amutha Prabakar Muniyandi, Daniel Arockiam, Feslin Anish Mon, N. Deepa · 2024
Federated Learning (FL) has emerged as a transformative approach for decentralized machine learning techniques, enabling collaborative model training while preserving data privacy. However, the distributed nature of FL imposes significant energy demands, particularly in resource-constrained environments. This chapter provides cutting-edge advancements in energy-efficient FL algorithms, focusing on methods that minimize computational and communication overhead without compromising model performance. It discusses key energy-intensive components in FL, including data transmission, local computation, and global aggregation, and investigates strategies such as model compression, sparse updates, adaptive aggregation, and client selection based on energy profiles. Additionally, it explores emerging techniques like quantized updates and knowledge distillation to optimize energy consumption. The chapter also highlights real-world applications and presents case studies that showcase trade-offs between energy efficiency, accuracy, and latency. Overall, it serves as a valuable resource for researchers aiming to enhance the sustainability of decentralized AI systems through a comprehensive overview of energy-efficient FL techniques.