Energy-Aware Federated Learning for Sustainable Wireless Edge AI

The Need

As federated learning expands across IoT, industrial, automotive, and connected-device ecosystems, maintaining reliable model training on energy-constrained edge devices remains a major challenge. Many devices operate on limited batteries or intermittent harvested energy, making continuous participation in distributed learning impractical. Existing federated learning systems often rely on centralized scheduling approaches that overlook device energy availability, communication conditions, and relative contribution to model performance, resulting in inefficient resource usage, slower convergence, and reduced learning accuracy in large-scale wireless environments.

The Technology

OSU engineers have developed an energy-aware framework for over-the-air federated learning that intelligently coordinates participation among distributed edge devices. The approach enables devices to autonomously determine when and how extensively to participate in model training based on their expected value to global model improvement and their available energy resources. A lightweight coordination mechanism helps maintain network-wide learning reliability while minimizing communication overhead. The result is a scalable and sustainable learning architecture that supports accurate model training across large wireless networks operating under realistic constraints such as intermittent energy availability, dynamic channel conditions, and heterogeneous data sources.

Commercial Applications

Energy-harvesting Internet of Things (IoT) and smart infrastructure networks
Industrial automation and predictive maintenance systems using battery-constrained sensors
Connected healthcare, wearable monitoring, and remote patient monitoring platforms
Vehicle-to-everything (V2X), intelligent transportation, and next-generation 6G edge AI systems