Dynamic Power Allocation in NOMA-Based Federated Learning System

Abstract

With the rapid rise of digital technology, artificial intelligence driven by big data has entered the fast lane of development, but it has also given rise to many problems, such as data silos and user privacy. A solution to solve these problems is federated learning. However, this framework also faces many challenges, with high communication costs in the first place. Non-orthogonal multiple access (NOMA) can be applied to alleviate the problem. In this paper, we focus on this issue and investigate multiple access technology based on federated learning. We build a NOMA-based federated learning system to improve the communication efficiency of federated learning. Then we propose a NOMA dynamic power allocation algorithm based on the realtime channel state at the edge user to improve the performance of the system. Experimental results show that the proposed algorithm can improve the training accuracy of the system model and reduce the energy consumption for uploading parameters.