Client Scheduling in Wireless Federated Learning Based on Channel and Learning Qualities
Jichao Leng, Zihuai Lin, Ming Ding, Peng Wang, David B. Smith, Branka Vucetic
Abstract
Federated learning (FL) emerges as a distributed training method in the Internet of Things (IoT), allowing participating clients to use their local data to train local models and upload parameters for global model aggregation after every few local iterations, protecting data privacy and reducing communication overhead. Given the scarcity of wireless communication resources, in this letter, we propose a client scheduling strategy for a wireless FL network based on a joint quality of channel and learning. Finally, we compare the proposed scheduling method’s performance with that of traditional methods considering the channel quality only. Experimental results show that our method can significantly improve training performance in terms of model accuracy and speed of convergence.