Energy-Efficient Distributed Machine Learning in Edge Computing Architectures

Abstract

The rapid growth of IoT devices, smart sensors, and real-time intelligent applications has increased the demand for distributed machine learning (DML) in edge computing environments. Traditional cloud-based machine learning approaches often suffer from high latency, excessive bandwidth consumption, privacy concerns, and increased energy costs. Edge computing addresses these challenges by enabling data processing closer to data sources, thereby reducing response time and network congestion. However, deploying machine learning on resource-constrained edge devices introduces challenges related to energy efficiency, computational limitations, communication overhead, and model synchronization. This study investigates energy-efficient distributed machine learning techniques, including federated learning, model compression, adaptive resource management, dynamic task offloading, and communication-efficient optimization. A distributed learning framework is proposed that integrates local model training, adaptive communication scheduling, gradient compression, and workload balancing to minimize energy consumption while maintaining learning accuracy. Mathematical models are developed to evaluate energy usage, communication costs, and convergence behavior. Experimental results demonstrate significant improvements in energy efficiency, network utilization, convergence speed, and resource management compared with conventional distributed learning approaches. The findings highlight the importance of intelligent communication reduction and adaptive edge orchestration for sustainable AI deployment. The proposed framework supports scalable, privacy-preserving, and real-time analytics in large-scale IoT environments, contributing to the development of next-generation energy-efficient edge intelligence systems.