MLWiNS: RL-based Self-driving Wireless Network Management System for QoE Optimization

MLWiNS：基于强化学习的自动驱动无线网络管理系统，用于 QoE 优化

• 批准号: 2003257
• 负责人: Arpit Gupta
• 金额: $ 56万
• 依托单位: University of California-Santa Barbara
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-08-01 至 2024-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2003257&HistoricalAwards=false
• 关键词: MLWiNS; RL; based; Self; driving

The proliferation of Internet-connected devices and applications, which are heavily reliant on high-speed, robust networks, have made the management of networks difficult and complicated. While the incorporation of programmable switches and machine-learning tools in data-centers and wide-area networks has contributed to improvements in Quality of Experience (QoE), access networks, which heavily rely on human operators for most network management tasks, remain the final frontier for providing a high quality of experience to end-users. The goal of this project is to design, implement, and evaluate RELIEF, a REinforcement-LearnIng (RL)-based sElF-driving wireless network-management system that can dynamically update network configurations to detect, diagnose, and resolve network events that contribute to QoE degradation. This project will develop new data-collection tools to capture fine-grained network data, new metrics to quantify QoE for the network, a scalable and explainable RL-based learning model to dynamically configure the underlying network and data-processing pipelines for QoE optimization. This project pursues fundamental contributions to the reinforcement learning area by developing novel solutions to inherent statistical and computational challenges. Finally, this project will design new programming abstractions and data structures to build an end-to-end system that optimizes the use of limited network resources to execute the trained RL models at scale.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

互联网连接的设备和应用程序的扩散非常依赖于高速，强大的网络，这使网络的管理变得困难且复杂。虽然将可编程开关和机器学习工具合并到数据中心和广阔的区域网络中，这有助于改善体验质量（QOE），访问网络，这些网络在大多数网络管理任务中都依赖人类运营商，但仍然是为最终用户提供高质量体验的最终领域。该项目的目的是设计，实施和评估救济，基于加强学习（RL）的自动驾驶无线网络管理系统，该系统可以动态更新网络配置以检测，诊断和解决导致QoE降级的网络事件。该项目将开发新的数据收集工具来捕获细粒网络数据，新的指标，以量化网络的QoE，这是一种可扩展且可解释的基于RL的学习模型，以动态配置基础网络和QoE优化的数据处理管道。该项目通过为固有的统计和计算挑战开发新颖的解决方案来追求对强化学习领域的基本贡献。最后，该项目将设计新的编程摘要和数据结构，以建立一个端到端系统，该系统优化了有限的网络资源以规模执行训练有素的RL模型。该奖项反映了NSF的法定任务，并被认为是值得通过基金会的知识分子优点和更广泛影响的审查标准通过评估来获得支持的。

项目成果

Estimating WebRTC Video QoE Metrics Without Using Application Headers

• DOI: 10.1145/3618257.3624828
• 发表时间: 2023-06
• 期刊: Proceedings of the 2023 ACM on Internet Measurement Conference
• 作者: Taveesh Sharma;Tarun Mangla;Arpit Gupta;Junchen Jiang;N. Feamster

Detecting Ephemeral Optical Events with OpTel

• 发表时间: 2022
• 作者: Congcong Miao;Minggang Chen;Arpit Gupta;Zili Meng;Lianjin Ye;Jingyu Xiao;Jie Chen;Zekun He;Xulong Luo;Jilong Wang;Heng Yu

Towards client-side active measurements without application control

实现无需应用程序控制的客户端主动测量

• DOI: 10.1145/3523230.3523234
• 发表时间: 2022
• 期刊: ACM SIGCOMM Computer Communication Review
• 影响因子: 2.8
• 作者: Goenka, Palak;Zarifis, Kyriakos;Gupta, Arpit;Calder, Matt
• 通讯作者: Calder, Matt

A NetAI Manifesto (Part I): Less Explorimentation, More Science

NetAI 宣言（第一部分）：更少的探索，更多的科学

• DOI: 10.1145/3626570.3626609
• 发表时间: 2023
• 期刊: ACM SIGMETRICS Performance Evaluation Review
• 作者: Willinger, Walter;Gupta, Arpit;Jacobs, Arthur S.;Beltiukov, Roman;Ferreira, Ronaldo A.;Granville, Lisandro
• 通讯作者: Granville, Lisandro

Panakos: Chasing the Tails for Multidimensional Data Streams

• DOI: 10.14778/3583140.3583147
• 发表时间: 2023-02
• 期刊: Proc. VLDB Endow.
• 作者: Fuheng Zhao;Punnal Ismail Khan;D. Agrawal;A. E. Abbadi;Arpit Gupta;Zaoxing Liu