Joint Bandwidth Allocation, Computation Control, and Device Scheduling for Federated Learning with Energy Harvesting Devices

Federated learning (FL) is a promising technique for distilling artificial intelligence from massive data distributed in Internet-of-Things (IoT) networks while keeping their privacy. However, the efficient deployment of FL faces several challenges due to e.g., limited radio resources, computation capability, and battery capacity of IoT devices. To address these challenges, in this work, the energy harvesting technique is first enabled on IoT devices for supporting their long-term training. Then, the convergence rate of the FL algorithm is derived, which indicates that for reducing the learning latency, the data utility, i.e., the number of training samples, should be maximized in each training iteration. To this end, a data utility maximization problem for each iteration is formulated, under the constraints of limited time, bandwidth, computation frequency, and energy supply. The problem is mixed-integer and non-convex, and hence NP-hard. A joint bandwidth allocation, computation control, and device selection scheme is proposed. In the scheme, an energy-efficient training data contribution indicator is first derived for each device, and then a sequential device scheduling scheme is designed.