Work

Spectrally Enhanced Policy Gradient Reinforcement Learning for Maximizing Connectivity Robustness in Dynamic Communication Networks

Year: 2026

Type: article

Abstract: Distributed control strategies have been employed as the fundamental basis for resource management and performance optimization in dynamic mobile edge networks, such as those found in urban smart cities and IoT systems. In these environments, traditional methods struggle to capture global connectivity dynamics, leading to suboptimal performance under rapidly changing network topologies. In this article, we propose a spectrally enhanced policy gradient reinforcement learning (SE-PGRL) scheme to address these challenges by integrating real-time spectral feedback into the control optimization process. Our approach leverages advanced graph-theoretic sensitivity analysis and iterative gradient-based updates with strong convergence guarantees to enhance state tracking accuracy, reduce control effort, and improve network resilience. Experimental results demonstrate that SE-PGRL achieves significant improvements, reducing mean squared error (mse) by 50%, decreasing control effort by 40%, and increasing connectivity robustness (CR) by 50%, thereby confirming its superiority over conventional methods.

Source: IEEE Internet of Things Journal

DOI: 10.1109/jiot.2026.3707518

Authors: Yang Lu, Yuhong Zhang, Yan Zheng, Wei Xiang

Institutions: Northeast Forestry University, University College London, Harbin Normal University, La Trobe University

Language: English

Topic: IoT and Edge/Fog Computing

Subfield: Computer Networks and Communications

Field: Computer Science

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: hybrid

Estimated APC paid: $2,645