A SER-based Device Selection Mechanism in Multi-bits Quantization Federated Learning

TL;DR

This work tackles the impact of wireless SER on federated learning by integrating non-orthogonal multiple access (NOMA) with MMSE-SIC and multi-bit gradient quantization. It introduces an inclusive SER-based device selection mechanism (SER-DSM) that allows many devices to participate while curbing the influence of poorly connected ones, and provides a global aggregation rule that accounts for SER and device eligibility. A theoretical convergence bound is derived, showing the FL performance upper bound $\mathbb{E}[F(\boldsymbol{w}^{[n+1]}) - F(\boldsymbol{w}^*)]$ depends on the convergence factor $A$, the SER-informed term $\Xi_k$, data sizes $\mathcal{D}_k$, and selection indicators, with convergence guaranteed when $A<1$. Empirical results on MNIST and Fashion-MNIST demonstrate that multi-bit gradient quantization and SER-DSM improve learning performance and convergence, identifying task- and system-dependent optimal quantization and modulation settings for robust FL over wireless links.

Abstract

The quality of wireless communication will directly affect the performance of federated learning (FL), so this paper analyze the influence of wireless communication on FL through symbol error rate (SER). In FL system, non-orthogonal multiple access (NOMA) can be used as the basic communication framework to reduce the communication congestion and interference caused by multiple users, which takes advantage of the superposition characteristics of wireless channels. The Minimum Mean Square Error (MMSE) based serial interference cancellation (SIC) technology is used to recover the gradient of each terminal node one by one at the receiving end. In this paper, the gradient parameters are quantized into multiple bits to retain more gradient information to the maximum extent and to improve the tolerance of transmission errors. On this basis, we designed the SER-based device selection mechanism (SER-DSM) to ensure that the learning performance is not affected by users with bad communication conditions, while accommodating as many users as possible to participate in the learning process, which is inclusive to a certain extent. The experiments show the influence of multi-bit quantization of gradient on FL and the necessity and superiority of the proposed SER-based device selection mechanism.

Figures (6)

• Figure 1: Federated learning under the NOMA computation framework.
• Figure 2: Participants Distribution in Simulation Experiment.
• Figure 3: MNIST (q=21840)
• Figure 4: Fashion-MNIST (q=28938)
• Figure 6: MNIST (q=21840)