Federated learning framework with battery-aware clients

Abstract

In this work, we propose BatteryFL, a novel Federated Learning (FL) framework that considers the energy constraints of clients and the ongoing data collection process. BatteryFL introduces a battery-aware client algorithm that enables clients to make informed decisions regarding their energy allocation during the FL process. Additionally, we present a server-side client selection algorithm that optimizes fairness and training loss objectives while considering the energy constraints of clients. We evaluate the proposed framework through extensive simulations, demonstrating the effectiveness of the BatteryFL framework in prolonging the lifetime of battery-powered clients while maintaining high learning performance.