Device Selection Methods in Federated Learning: A Survey

Federated learning, a type of distributed machine learning, is used as local learning on large, distributed datasets produced by massive devices. It eliminates the need for data to be transported from end devices to a cloud or central server. In order to aggregate global models, it entails learning local datasets using models on devices and passing the learned parameters to a central server. The primary difficulty faced in federated learning for training is identifying devices at a given time. This paper primarily focuses on the analysis of the selection methodology for identifying the optimal sensor or device for training using federated learning algorithms. To make well-informed judgements, factors including device capabilities, network circumstances, privacy restrictions, and data quality are taken into account. The research indicates that the most suitable sensor or device should be dynamically chosen for each training iteration in order to optimize the federated learning process’s accuracy and performance. Utilizing the traditional learning patterns of supervised, unsupervised, and reinforcement learning, a comparative analysis of the three approaches is presented in the paper. This comparison includes the efficacy and efficiency of the approaches, demonstrating enhanced training accuracy and resource utilization in different environments.