FedCLS:A federated learning client selection algorithm based on cluster label information

Federated learning is a common distributed machine learning framework. Through the training of the global model, the problems of large communication overhead and data privacy protection in traditional centralized machine learning are solved. But in real distributed scenarios, Non-Independent and Identically Distributed(Non-IID) of data reduces the speed of learning and the accuracy of global model. To solve this problem, this paper proposes a federated learning client selection algorithm based on cluster label information(FedCLS). FedCLS realizes efficient federated learning by optimizing the selection of clients in each round of training. Through extensive simulations, we demonstrate that compared with traditional FedAVG based on random extraction, FedCLS has better learning performance and less resource overhead.