Adaptive Federated Reinforcement Learning for Network Management in Complex Communication Networks

Modern communication networks face dynamic conditions from video streaming, Augmented Reality/Virtual Reality (AR/VR), and Internet of Things (IoT) devices. We propose a federated multi-agent reinforcement learning (F-MARL) framework that leverages adaptive multi-path routing for congestion control while preserving privacy through federated coordination. Our event-driven approach triggers Federated Learning (FL) only under congestion, with energy-aware and Shapley-based client selection for resource-constrained devices. Implemented in OMNeT++, the framework achieves 28% congestion reduction versus Open Shortest Path First (OSPF), 15% throughput improvement over Multi-Agent Congestion Control (MACC), and 42% lower overhead than non-federated Multi-Agent Reinforcement Learning (MARL). Results demonstrate privacy-preserving, adaptive intelligence for next-generation networks.