Distributed Network Intelligence over Mobile Edge Systems

Network intelligence is envisioned as the cornerstone of 6G mobile edge systems. By deploying distributed machine learning algorithms at end-user terminals, the network can harness collective intelligence from vast data sources, enabling the real-time management of numerous devices in dynamic environments. However, users’ mobility nature renders the end-user terminals constantly encountering varying physical conditions, necessitating fast convergence of the distributed model training for timely decision-making. To accomplish this target, this project aims to deliver a systematic design, grounded in solid theoretical foundations and validated through real-world prototypes, that advances the understanding, optimization, and implementation of distributed network intelligence in mobile edge systems. The objectives are fourfold: (1) establish a theoretical framework for assessing the impact of system parameters across different layers on the efficiency of model training and identifying the dominating factors that constrain the training performance; (2) develop a goal-oriented status information exchange mechanism to enable timely update of status information across the interconnected terminals; (3) design a novel distributed model training method, utilizing personalized federated learning and high-order distributed optimization methods to stabilize the model update and accelerate the convergence process; and (4) build a comprehensive prototype system that implements the proposed solutions and verify their practical efficacy. These innovations will not only enhance the interpretability of network intelligence but also improve its operability. The expected outcome of this project shall forge forward the studies in distributed machine learning as well as network communication theory, demonstrating that the innovative fusion of two traditionally hitherto unrelated fields of study will propel and facilitate the rapid development and deployment of network intelligence in 6G.