Optimizing Secrecy Rate for Federated Learning Model Aggregation With Intelligent Reflecting Surface Toward 6G Ubiquitous Intelligence

Abstract

Non-Orthogonal Multiple Access (NOMA) based Federated Learning (FL) can achieve the massive connectivity of Internet of Thing (IoT) devices, high transmission rate, and pervasive intelligence in 6G networks. However, the stochastic channels and frequent model parameter updates may incur degraded transmission rate and diminished FL performance, while privacy leakage may happen if Eavesdroppers (Eves) intercept the FL training process. To address the above issues, we exploit Intelligent Reflecting Surface (IRS) to reconfigure wireless signal propagation for secure transmission and fast convergence of NOMA-based FL. In this article, a Deep Reinforcement Learning (DRL) based approach is proposed to jointly optimize the transmission power of edge devices and IRS phase shift to maximize the minimum secrecy rate in the model parameter uploading process. Numerical results validate the efficiency of our proposed algorithm and demonstrate that IRS can improve the secrecy rate.