A Priority-Aware Task Offloading Method with Knowledge Distillation in Edge Computing

With the swift proliferation of Internet of Things (IoT) technology, offloading tasks efficiently is essential for enhancing system performance. Numerous strategies for task offloading in edge computing environments have been introduced. They often fall short of effectively managing task prioritization and latency requirements. To address this issue, this paper introduces a priority-aware task offloading approach that integrates Deep Reinforcement Learning (DRL) with Knowledge Distillation (KD). The method employs a Dueling duDeep QNetwork (Dueling DQN) model coupled with a priority-weighted reward mechanism to refine task offloading decisions. By incorporating a pre-trained teacher model, we facilitate the learning process of the DRL agent and bolster its performance. The experimental outcomes demonstrate that our proposed approach outperforms four comparative benchmark methods, achieving notable improvements in average latency, the ratio of overdue tasks, and the priority-weighted average delay.