FedSAR for Heterogeneous Federated learning：A Client Selection Algorithm Based on SARSA

Federated learning is an emerging distributed machine learning paradigm. It only uploads model parameters while keeping data locally, effectively protecting user privacy and attracting widespread attention due to its great potential. However, there are significant differences in data distribution, model architecture, and hardware devices among the client devices participating in the training process, which may greatly affect the data accuracy and aggregation speed of the model. To reduce the interference caused by model heterogeneity and data heterogeneity on the data accuracy and aggregation speed of federated learning, this paper proposes a client selection method based on the SARSA algorithm for heterogeneous federated learning. Firstly, K client devices are selected to participate in the federated training through the SARSA algorithm. After the client devices complete the local training, the central server aggregates the model parameters of the K client devices and distributes the updated parameters to the client devices. Experiments demonstrate that the algorithm proposed in this paper can effectively improve the model performance, enhance communication efficiency, and accelerate model convergence.