Abstract
Internet of things system is generating huge amounts of data all the time, and the accompanying challenge is the difficulty of real-time processing and analysis. In this paper, we propose a real-time subspace learning framework based on federated learning (RTFed), in which participating devices communicate in over-air mode. RTFed reduces the influence of missing data and outliers on distributed learning by using subspace tracking method under the condition of data sharing constraints. Moreover, the proposed algorithm effectively alleviates the common channel noise problem in over-air broadcasting by using compressive sensing method. In addition, RTFed uses the multi-core parallelism of CPUs and GPUs to enhance the processing of real-time data, ensuring timely updates for applications such as recommendation systems and social networks. The performance of the framework was demonstrated through extensive numerical experiments and real-world data evaluation.