A Multi-Agent Meta-Based Task Offloading Strategy for Mobile Edge Computing

Task offloading in mobile edge computing (MEC) improves the efficacy of devices terms performance, data storage, and energy consumption by computational tasks to servers. Efficient task can leverage MEC technology reduce processing latency consumption. By integrating reasoning ability machine intelligence cognitive architecture, such as SOAR ACT-R, reinforcement learning (RL) algorithms have been applied resolve MEC. To solve problem that conventional Deep RL (DRL) cannot adapt dynamic environments, this paper proposed a scheduling strategy which combined multi-agent meta-learning. In order make two actions charging time fully considered at same time, we implemented network agent on device. efficiently train policy network, first approximation method based clipped surrogate objective. Finally, experiments are designed with variety number subtasks, transmission rate, server results show MRL-based has overwhelming overall performance be quickly various environments good stability generalization.