Selective Federated Learning for Mobile Edge Intelligence
Xin Yuan, Ke Zhang, Yan Zhang · 2021 13th International Conference on Wireless Communications and Signal Processing (WCSP) · 2021
Federated learning enables distributed agents to train a common and shared model in a privacy-protected way. Due to the heterogeneity of computing and communication capabilities, these agents may produce different training effects and asynchronous model update in the learning process. To improve the learning efficiency, here emerges the challenge to optimize agents' selection and collaboration. Catering for this challenge, in this paper, we leverage Shapley Value to measure the learning contribution of each agent in the model training. Moreover, we propose joint optimization schemes of agent selection and resource scheduling that maximize learning energy efficiency. Numerical results demonstrate that our schemes outperform benchmark schemes in various data distribution scenarios.