.png){kind=icon}
返回
TayMAML: A meta reinforcement learning-based task scheduling method for edge computing
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2026.131253.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
This paper presents TayMAML, an edge computing task scheduling algorithm designed to address the challenges of suboptimal generalization and the trade-off between computational efficiency and accuracy in traditional meta-reinforcement learning algorithms within dynamically heterogeneous edge environments. To enhance task scheduling performance, we first propose a biased sampling strategy that evaluates task learning progress based on training loss. This strategy determines the number of test samples for various tasks, ensuring consistency between training and testing task distributions. Additionally, a lightweight distribution consistency strategy is introduced to further reduce disparities between training and testing distributions. This approach quantifies distribution differences and incorporates these differentials into the original meta-loss for meta-updates. Through theoretical derivation, we isolate the second-order derivative term in the meta-update process. Leveraging Taylor expansion, we derive a first-order approximation of the second-order derivative, enabling precise parameter updates while avoiding the computational overhead typically associated with second-order derivatives in meta-reinforcement learning. Experimental evaluations demonstrate that TayMAML significantly improves model generalization and stability, reduces system latency and energy consumption, and effectively supports real-time task requirements in dynamically heterogeneous edge environments, outperforming existing state-of-the-art algorithms.
期刊
IF:
7.5
论文数:
2.9W
被引数:
10.2W
机构
引用论文
A comprehensive survey on reinforcement-learning-based computation offloading techniques in Edge Computing Systems边缘计算系统中基于强化学习的计算卸载技术综述
OAAI
OAAI分享收藏Collaborative cloud-edge-end task offloading with task dependency based on deep reinforcement learning基于深度强化学习的协同任务依赖的云-边端任务卸载
PREAI
分享收藏Deep Learning-Based Dynamic Computation Task Offloading for Mobile Edge Computing Networks基于深度学习的移动边缘计算网络动态计算任务卸载.png)
SENSORSOAAI
SENSORSIF3.5
OAAI分享收藏Deep Reinforcement Learning for Energy-Efficient Computation Offloading in Mobile-Edge Computing面向移动边缘计算节能计算卸载的深度强化学习
PREAI
分享收藏
查看更多内容