Energy-Efficient Federated Knowledge Distillation Learning in Internet of Drones

Federated Learning (FL) in the Internet of Drones (IoD) leverages the distributed computational resources of drones for collaborative learning, while addressing challenges such as data privacy and limited bandwidth in aerial networks. Federated Knowledge Distillation (FedKD) addresses the challenges of FL in IoD by reducing the model size and communication overhead, thus enabling more effective and scalable machine learning across drone networks. This paper investigates energy-efficient FedKD in IoD networks and focuses on optimizing CPU frequencies, a key factor in reducing energy consumption during the learning process. We aim to minimize the overall energy use of drones while meeting strict latency requirements for training, optimizing CPU frequencies for both teacher and student models. This problem is formulated as a non-linear programming problem, and we introduce an efficient algorithm to address it. Extensive simulations are conducted to validate the performance of our proposed algorithm.