Scholarlink

<p>Ensuring optimal and timely task offloading in dynamic mobile edge computing (MEC) environments is critical yet challenging. Although meta-reinforcement learning (MRL) can achieve offloading objectives with limited samples, it fails to adapt effectively to fluctuations in task complexity or distinct learning stages. To overcome this rigidity, this paper proposes an adaptive learning rate meta-reinforcement learning (ALR-MRL) algorithm. First, to mitigate resource constraints on edge nodes, we incorporate cloud servers into the offloading architecture, thereby significantly enhancing system adaptability and computational capacity. Second, we develop a novel policy update mechanism combining Seq2Seq networks with adaptive learning rates. By converting task graphs into embedding sequences and dynamically adjusting learning rates based on gradient variance and bias, the algorithm achieves precise adaptation to varying task difficulties. Finally, we validate the proposed method through comprehensive simulations. The results show that ALR-MRL outperforms existing methods, reducing average latency by 30% relative to baselines and by 40% relative to fine-tuning deep reinforcement learning algorithms.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic task offloading strategy in mobile edge computing using meta-reinforcement learning with adaptive learning rate adjustment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月14日

查看全文
  • Rui Yao,
  • Yong Wang,
  • Junyan Chen,
  • Lei Jin,
  • Jiahao Chen

摘要

Ensuring optimal and timely task offloading in dynamic mobile edge computing (MEC) environments is critical yet challenging. Although meta-reinforcement learning (MRL) can achieve offloading objectives with limited samples, it fails to adapt effectively to fluctuations in task complexity or distinct learning stages. To overcome this rigidity, this paper proposes an adaptive learning rate meta-reinforcement learning (ALR-MRL) algorithm. First, to mitigate resource constraints on edge nodes, we incorporate cloud servers into the offloading architecture, thereby significantly enhancing system adaptability and computational capacity. Second, we develop a novel policy update mechanism combining Seq2Seq networks with adaptive learning rates. By converting task graphs into embedding sequences and dynamically adjusting learning rates based on gradient variance and bias, the algorithm achieves precise adaptation to varying task difficulties. Finally, we validate the proposed method through comprehensive simulations. The results show that ALR-MRL outperforms existing methods, reducing average latency by 30% relative to baselines and by 40% relative to fine-tuning deep reinforcement learning algorithms.