Optimal Client Selection for Wireless Federated Learning

In order to address the issue of statistical heterogeneity in wireless federated learning, we proposed a novel client selection method. We introduced a new metric called client quality, which takes into account the local data volume of clients and the testing accuracy of the previous round's model. Based on this, we proposed a client selection method called Client Quality-based Client Selection (CQCS) to improve the performance of wireless federated learning. Results show that the proposed CQCS client selection method effectively enhances the global accuracy of the model. Specifically, compared to the baseline algorithms FedAvg, FedProx, and Per-FedAvg, the model average accuracy improved by 6.7%, 4.8%, and 3.1%, respectively, on the MNIST dataset. On the Cifar10 dataset, the model average accuracy improved by 3.4 %,3.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>, and 3.9%, respectively. Similarly, on the Fashion-MNIST dataset, the model average accuracy improved by 3.1 %, 3.8%, and 4.2%, respectively.