"Graph-Augmented MAPPO: Dynamic Topology-Aware Task Offloading Optimization in Edge Computing's data"

"In multi-access edge computing (MEC) environments, joint optimization of task offloading and resource allocation is challenging due to dynamic topologies, limited resources, and network heterogeneity. This letter proposes a Graph-Augmented Multi-Agent Proximal Policy Optimization (GA-MAPPO) framework under a centralized training and decentralized execution (CTDE) paradigm. A Master\u2013Agent cooperative architecture integrates a Graph Attention Network (GAT) to model dynamic connectivity and resource variations. The method employs a hybrid action space for unified optimization of offloading and resource allocation. Extensive simulations with real-world mobility and workload traces show that GA-MAPPO outperforms representative baselines in task completion rate, energy efficiency, latency, and training stability, demonstrating its effectiveness and practical potential in dynamic MEC and Internet of Vehicles (IoV) scenarios."

在多接入边缘计算（Multi-Access Edge Computing, MEC）环境中，由于拓扑动态变化、资源受限以及网络异构性，任务卸载与资源分配的联合优化极具挑战性。本快报提出一种基于集中式训练、分布式执行（Centralized Training and Decentralized Execution, CTDE）范式的图增强型多代理近端策略优化（Graph-Augmented Multi-Agent Proximal Policy Optimization, GA-MAPPO）框架。该框架采用主-代理协同架构，集成图注意力网络（Graph Attention Network, GAT）以建模动态连通性与资源变化情况。所提方法采用混合动作空间，实现任务卸载与资源分配的统一优化。基于真实世界移动性与工作负载轨迹的大规模仿真实验表明，GA-MAPPO在任务完成率、能源效率、时延及训练稳定性方面均优于代表性基准方法，证实了其在动态MEC与车联网（Internet of Vehicles, IoV）场景中的有效性与实用潜力。