Joint Optimization of Fairness and Energy Efficiency in Zero-Trust Federated Learning for Consumer Internet of Things: A Lossy Communication Perspective

Abstract

The Consumer Internet of Things (CIoT) represents a transformative technological paradigm, seamlessly integrating the digital and physical worlds to enhance and simplify everyday life. In this context, Zero-Trust Federated Learning (ZTFL) further empowers CIoT and spawns a series of emerging applications. However, zero-trust federated learning often encounters bottlenecks in energy consumption and efficiency. The above bottleneck problems hinder the further application and expansion of CIoT, which is not conducive to the healthy development of CIoT. Therefore, this study focuses on the issue of joint optimization of fairness and energy efficiency in ZTFL, particularly in the context of lossy communications. This research explores the fairness problem of ZTFL by Lyapunov optimization in a frequency division multiple access (FDMA) system to address these challenges. It proposes a method of dynamic queue quantification of consumer electronics device participation that considers the packet loss rate. Analyzing the objective function, we transform some non-convex functions into convex functions and provide analytical solutions. Furthermore, we experimentally evaluate our model using the MNIST, Cifar10, and Cifar100 datasets, with results showing that, under lossy communications, our proposed model can significantly improve model accuracy while maintaining an average 26% reduction in communication completion. Our data and code are available at https://github.com/sunjia123456789/The-Faired-FL.