Energy-Efficient Federated Learning for Distributed Edge–Cloud Computing Environments

Abstract

Federated learning enables distributed model training across edge devices and cloud resources without transferring raw data to a central repository. However, the energy implications of periodic communication, heterogeneous hardware, unbalanced participation, and continuous model aggregation remain insufficiently addressed in production edge-cloud environments. This paper presents a system-level analysis of energy-efficient federated learning, emphasizing architectural trade-offs, resource governance, deployment policy, and sustainability. The central argument is that energy efficiency cannot be achieved solely through algorithmic compression or hardware acceleration; it must be designed as an emergent property of the interaction among protocol design, edge-cloud topology, operational incentives, and fairness constraints. The paper examines the structural differences between centralized and hierarchical federated architectures, explores communication-computation trade-offs, and discusses how serverless cloud orchestration and edge resource management affect energy consumption. It further addresses robustness and fairness, showing that energy optimization strategies may create uneven participation burdens on battery-powered devices and underrepresented nodes. Governance frameworks are considered through the lens of shared infrastructure management, carbon accounting, and multi-stakeholder coordination. The analysis connects federated learning to broader sustainable machine learning debates and offers forward-looking research directions in carbon-aware scheduling, energy reporting standards, and policy-aware system design. Rather than proposing a single optimization algorithm, the paper contributes a conceptual foundation for designing federated learning infrastructures that balance training quality, privacy, equity, and environmental responsibility.