ALFL: A Federated Learning Client Selection Algorithm for Heterogeneous Data
Guan Wang, Runyang Liu, Shihao Liu · 2024
Data heterogeneity of clients is one of the main factors affecting the performance of federated learning. Since the data distribution on each client has large differences, which makes different clients participate in the aggregation of federated learning with different effects, the selection strategies of federated learning clients are a hot topic in recent years. However, they have less considered the impact of active learning strategies in federated learning. In this paper, we propose a method to select clients using active learning strategies to address the impact of data heterogeneity on federation learning and improve the accuracy of the model. Specifically, during each selection of clients in federated learning, a portion of clients are selected to participate in aggregation through some active learning strategies.We empirically evaluate our ALFL method on the publicly available datasets FMNIST, CIFAR-10 and LEAF-synthetic dataset. The experimental results show that the method proposed in this paper can significantly reduce the number of federated learning communication rounds and improve model accuracy.