Research Article
Distributed Unsupervised Learning-Based Task Offloading for Mobile Edge Computing Systems
@INPROCEEDINGS{10.1007/978-3-030-67720-6_37, author={Jianming Wei and Qiuming Liu and Shumin Liu and Yiping Zeng and Xin Xiong}, title={Distributed Unsupervised Learning-Based Task Offloading for Mobile Edge Computing Systems}, proceedings={Communications and Networking. 15th EAI International Conference, ChinaCom 2020, Shanghai, China, November 20-21, 2020, Proceedings}, proceedings_a={CHINACOM}, year={2021}, month={2}, keywords={Mobile edge computing Distributed unsupervised learning Energy efficiency Task offloading}, doi={10.1007/978-3-030-67720-6_37} }- Jianming Wei
Qiuming Liu
Shumin Liu
Yiping Zeng
Xin Xiong
Year: 2021
Distributed Unsupervised Learning-Based Task Offloading for Mobile Edge Computing Systems
CHINACOM
Springer
DOI: 10.1007/978-3-030-67720-6_37
Abstract
Mobile edge computing has emerged as a new paradigm to enhance computing capabilities by offloading complicated tasks to nearby cloud server. To conserve energy as well as maintain quality of service, algorithms with low time complexity for task offloading is required. In this paper, a multi-user with multiple tasks scenario is considered, taking full account of factors including data size, bandwidth, channel state information, we propose a distributed unsupervised learning-based offloading algorithm for task offloading, where distributed parallel networks are employed to guarantee the robustness of algorithm. Additionally, we exploit a memory pool to store input data and corresponding decisions as key-value pairs. Based on the experience mechanism, the proposed algorithm can omit the step of data calibration compared with supervised learning method. To further reduce the communication cost, we analyze four bandwidth allocation schemes. Results reveal that the channel state-based strategy cost 3% less than data size-based, mixed and user size-based. Besides, the proposed algorithm can save 14, 18% cost than local and edge schemes respectively. Numerous results show that the proposed algorithm can achieve near-optimal decisions timely as well as having high reliability.
