Exploring Improved Asynchronous Federated Learning with Fresh Information in Contested Environment

To tackle weaknesses of traditional synchronous federated learning (FL), asynchronous FL has emerged as a solution. However, asynchronous FL also faces challenges, particularly in identifying optimal clients capable of contributing to a global model with minimal latency and loss with high accuracy, while prioritizing information freshness. This paper explores client selection strategies in asynchronous FL that takes account of computing and communication delays, freshness of the data/information being used in generating local models, and the loss while providing the best global model with high efficiency in the contested environment. Furthermore, we present formal mathematical analysis and performance evaluation using numerical results. Results show that the proposed improved asynchronous FL outperforms the other FL approaches.