[info] T. Benson, A. Akella, and A. Shaikh, “Demystifying configuration challenges and trade-offs in network-based ISP services,” in Proceedings of the ACM SIGCOMM 2011 conference, Toronto Ontario Canada: ACM, Aug. 2011, pp. 302–313. doi: 10.1145/2018436.2018471. [/info]

[info] C. Streiffer, H. Chen, T. Benson, and A. Kadav, “DeepConfig: Automating Data Center Network Topologies Management with Machine Learning.” arXiv, Dec. 11, 2017. Accessed: Aug. 06, 2023. [Online]. Available: http://arxiv.org/abs/1712.03890 [/info]

[info] H. Kim, T. Benson, A. Akella, and N. Feamster, “The evolution of network configuration: a tale of two campuses,” in Proceedings of the 2011 ACM SIGCOMM conference on Internet measurement conference, Berlin Germany: ACM, Nov. 2011, pp. 499–514. doi: 10.1145/2068816.2068863. [/info]

[info] T. Benson, A. Akella, and D. A. Maltz, “Mining policies from enterprise network configuration,” in Proceedings of the 9th ACM SIGCOMM conference on Internet measurement, Chicago Illinois USA: ACM, Nov. 2009, pp. 136–142. doi: 10.1145/1644893.1644909. [/info]

[info] T. Benson, A. Akella, and D. Maltz, “Unraveling the Complexity of Network Management,” in Proceedings of the 6th USENIX Symposium on Networked Systems Design and Implementation, in NSDI’09. USA: USENIX Association, 2009, pp. 335–348. doi: 10.5555/1558977.1559000. [/info]

1.1 背景

• 企业网络配置非常复杂，不同网络之间配置差异很大。
• 企业网络中，配置错误是造成网络故障的一个重要原因。

[info] B. Li et al., “Random sketch learning for deep neural networks in edge computing,” Nat Comput Sci, vol. 1, no. 3, pp. 221–228, Mar. 2021, doi: 10.1038/s43588-021-00039-6. [/info]

1.1 背景

深度神经网络对计算和存储资源需求巨大，这给它们在边缘设备上的部署带来困难。最近，轻量级深度学习受到了极大关注，其目的是通过网络剪枝、低秩近似(LRA)、权重量化和网络架构转换(NAT)等压缩大型DNN模型。有工作基于矩阵逼近理论近似相对更低秩和稀疏的DNN模型的权重矩阵，从而得到一个轻量的紧凑模型。

[info] R. G. Pacheco, K. Bochie, M. S. Gilbert, R. S. Couto, and M. E. M. Campista, “Towards Edge Computing Using Early-Exit Convolutional Neural Networks,” Information, vol. 12, no. 10, p. 431, Oct. 2021, doi: 10.3390/info12100431. [/info]

1.1 背景

在计算机视觉应用中,移动设备由于计算资源的限制往往需要将卷积神经网络(CNN)的推理任务转移到云端完成。但是除了给云端带来更多的网络负载之外,这种方法还会使那些需要低延迟的应用变得不可行。

[info] J. Yao et al., “Edge-Cloud Polarization and Collaboration: A Comprehensive Survey for AI,” IEEE Trans. Knowl. Data Eng., pp. 1–1, 2022, doi: 10.1109/TKDE.2022.3178211. [/info]

[info] E. Li, L. Zeng, Z. Zhou, and X. Chen, “Edge AI: On-Demand Accelerating Deep Neural Network Inference via Edge Computing,” IEEE Trans. Wireless Commun., vol. 19, no. 1, pp. 447–457, Jan. 2020, doi: 10.1109/TWC.2019.2946140. [/info]

1.1 背景

深度神经网络(DNN)是支持移动智能应用的关键技术，但在移动设备上运行DNN存在计算资源有限的挑战。传统的云计算辅助的DNN推理存在明显的延迟问题。边缘计算作为一种新兴的计算模式，可以用于支持实时的DNN推理。

[info] C. Hu and B. Li, “Distributed Inference with Deep Learning Models across Heterogeneous Edge Devices,” in IEEE INFOCOM 2022 - IEEE Conference on Computer Communications, London, United Kingdom: IEEE, May 2022, pp. 330–339. doi: 10.1109/INFOCOM48880.2022.9796896. [/info]

1.1 背景

随着深度学习在各种任务中的广泛应用,将模型部署到边缘设备进行推理非常重要。但是直接在资源受限的边缘设备上使用复杂的深度学习模型进行推理可能会很慢。因此,通过分布式推理来加速非常必要。