An adaptive asynchronous federated learning framework for heterogeneous Internet of things

Weidong Zhang, Dongshang Deng, Xuangou Wu, Wei Zhao, Zhi Liu, Tao Zhang, Jiawen Kang, Dusit Niyato

Abstract: Highlights•We propose an adaptive asynchronous federated learning framework to reduce communication overhead.•An adaptive client scheduling strategy is employed to assess the value of model parameters and control their upload.•An adaptive parameter aggregation method is employed to optimize server-side aggregation weights and accelerate convergence.•We theoretically prove the convergence of the method, and the results on multiple open-source datasets also show validity.