SAGE: Sustainable and green energy-aware client selection for federated learning in IoT

The Internet of Things (IoT) is widely adopted in industrial applications and smart infrastructures, enabling large-scale data collection and intelligent services. However, the distributed nature of IoT networks introduces significant energy challenges, as devices often operate under heterogeneous conditions and with highly variable energy availability. Ensuring the sustainability of such systems requires adaptive energy management strategies and localized processing techniques that reduce unnecessary energy usage while preserving reliable performance. Federated Learning (FL) has emerged as a promising paradigm for IoT, allowing the collaborative training of a global model on a server aggregating parameters from different devices, the clients, without exposing private data. Nevertheless, the repetitive communication and computation inherent to FL can exacerbate energy constraints. To minimize energy consumption, communication strategies such as model compression can be used to transmit updates, or client selection (CS), to select clients that can effectively improve the global model with their updates.