arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2104.07351cs.NIeess.SP
收藏

基于增强LSTM的移动增强现实服务分解

Enhanced LSTM-based Service Decomposition for Mobile Augmented Reality

Zhaohui Huang, Vasilis Friderikos

摘要:

毫无疑问,用于5G及更远无线网络的移动增强现实(MAR)应用最近受到了显著的关注。然而,它们需要通过边缘云(EC)支持在终端设备和/或网络上使用大量的计算和存储资源。在这项工作中,MAR服务被认为是在微服务的镜头下,MAR服务组件可以被分解和锚定在从终端设备到不同EC的不同位置,以优化整体服务和网络效率。为此,我们提出了一种移动性感知MAR服务分解,使用长短期记忆(LSTM)深度神经网络提供有效的主动决策实时。更具体地,LSTM深度神经网络是用从数学编程公式中以离线方式导出的最优解来训练的。然后,在推理阶段的决策是用来优化MAR服务的服务分解。大量的数值研究表明,移动性感知LSTM深度神经网络在决策质量和计算时间方面都优于最近提出的方案。

英文摘要:

Undoubtedly, Mobile Augmented Reality (MAR) applications for 5G and Beyond wireless networks are witnessing a notable attention recently. However, they require significant computational and storage resources at the end device and/or the network via Edge Cloud (EC) support. In this work, a MAR service is considered under the lenses of microservices where MAR service components can be decomposed and anchored at different locations ranging from the end device to different ECs in order to optimize the overall service and network efficiency. To this end, we propose a mobility aware MAR service decomposition using a Long Short Term Memory (LSTM) deep neural network to provide efficient pro-active decision making in real-time. More specifically, the LSTM deep neural network is trained with optimal solutions derived from a mathematical programming formulation in an offline manner. Then, decision making at the inference stage is used to optimize service decomposition of MAR services. A wide set of numerical investigations reveal that the mobility aware LSTM deep neural network manage to outperform recently proposed schemes in terms of both decision making quality as well as computational time.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑