A Multi-Agent Federated Edge Learning System for Energy-Efficient 6G Wireless Ecosystems

Abstract

In this article, we develop an Intelligent Federated Edge Learning Framework for enhancing energy efficiency as well as the performance of learning models within the context of a 6G Wireless Network. Specifically, our proposed Intelligent Federated Edge Learning Framework consists of combined adaptive client selection, energy aware aggregation and node participation using Reinforcement Learning approaches. Unlike conventional, Cloud based approaches, our new framework will reduce the amount of unnecessary communication rounds by only using optimal edge devices based on their channel quality and their remaining energy. The empirical test results indicate that the new system will reduce the amount of energy consumed by 32%, increase the rate at which models converge to the desired solution by 27%, and increase the overall network throughput by 22%, relative to models developed using either the baseline federated learning models or the baseline centralized learning models. Additionally, testing found that the latency is lowered by 25%, allowing for use in time-sensitive applications. In addition to these improvements, the proposed framework provides similar levels of accuracy as previous systems, while greatly reducing the amount of communication overhead associated with model training. The experimental findings indicate that the proposed approach can provide both energy-efficient and effective learning solutions, making it an ideal candidate for future large scale IoT systems and future generations of wireless networks.