RINGS: Object-Oriented Video Analytics for Next-Generation Mobile Environments

RINGS：下一代移动环境的面向对象视频分析

• 批准号: 2147909
• 负责人: Ravi Netravali
• 金额: $ 100万
• 依托单位: Princeton University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-05-01 至 2025-04-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2147909&HistoricalAwards=false
• 关键词: RINGS; Object; Oriented; Video; Analytics

Over the past few decades, cellular networks have evolved to deliver improved performance across increasingly heterogeneous components spanning the network edge (e.g., user devices) to base stations to traditional cloud backends. A key motivator behind these advances is to enhance the support for edge applications, especially video analysis (VA). Yet VA applications are currently not structured to fully leverage those advances. A primary issue is the lack of structured frameworks to develop and run VA applications, which in turn prevents the deployment and optimizations required to take advantage of all that cellular networks (and their edge-cloud hierarchies) have to offer. To tackle this limitation, the proposed work advocates for a re-designed VA software stack that explicitly ties VA operations and requirements to the resources, interfaces, and vantage points that each platform element in a mobile edge-cloud hierarchy brings. To achieve this goal, the project takes a bottom-up, three-pronged approach that involves (1) developing a new object-oriented query language for VA applications that makes the aforementioned characteristics explicit and observable, (2) leveraging those features to develop a suite of resource-aware optimizations to VA computations that can operate under diverse (and restricted) edge constraints, and (3) designing a novel task placement engine that automatically adapts and operates VA applications across edge-cloud hierarchies.Owing to the widespread use of VA applications in sectors spanning traffic control, to autonomous vehicles, to disaster relief, the proposed research promises benefits to a large part of the population. The key improvements will come along two axes – (1) replacing painstaking manual analysis with automatic determination of the appropriate interactions between VA applications and emerging mobile networking infrastructure, and (2) democratizing the use of edge networking infrastructure – and will target two different groups. On the one hand, the proposed frameworks will simplify the creation of cutting-edge VA applications for developers by automatically deciding what public edge infrastructure to use and how to use it most effectively (in terms of cost, accuracy, and performance). On the other hand, the developed systems will assist network operators in identifying the most fruitful resource enhancements and helpful information about the platform to expose to application elements. The project also involves outreach efforts to attract students from populations currently under-represented in computer science. Key to these efforts is magnifying the interdisciplinary nature of edge-based VA applications that span mobile systems and networks, computer vision, programming languages, and machine learning. The software and research artifacts designed as part of this project are released on a regularly-maintained, public website.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.

在过去的几十年里，蜂窝网络已经发展到可以在网络边缘（例如用户设备）、基站和传统云后端的日益异构的组件上提供更高的性能，这些进步背后的一个关键推动力是增强对边缘应用程序的支持。 ，尤其是视频分析 (VA)，但 VA 应用程序目前的结构尚未充分利用这些进步，一个主要问题是缺乏开发和运行 VA 应用程序的结构化框架，这反过来又阻碍了利用这些进步所需的部署和优化。所有蜂窝网络（及其边缘云层次结构）必须提供解决此限制的方案，建议重新设计 VA 软件堆栈，将 VA 操作和要求与每个平台元素所在的资源、接口和优势点明确联系起来。为了实现这一目标，该项目采用了自下而上的三管齐下的方法，其中包括（1）为 VA 应用程序开发一种新的面向对象的查询语言，使特征变得明确且可观察，（ 2）利用这些功能开发一套针对 VA 计算的资源感知优化，可以在不同（且受限）的边缘约束下运行，以及（3）设计一种新颖的任务放置引擎，可以跨边缘云层次结构自动适应和操作 VA 应用程序。通过在交通控制、自动驾驶汽车、救灾等领域使用 VA 应用，拟议的研究有望使大部分人口受益，关键改进将集中在两个方面：(1) 用自动确定取代繁琐的手动分析。之间适当的相互作用VA 应用程序和新兴的移动网络基础设施，以及 (2) 使边缘网络基础设施的使用民主化——并且将针对两个不同的群体。一方面，所提出的框架将通过自动决策来简化开发人员尖端 VA 应用程序的创建。另一方面，开发的系统将帮助网络运营商确定最富有成效的资源增强和有关平台的有用信息。该项目还涉及到外展工作。这些努力的关键是扩大基于边缘的 VA 应用程序的跨学科性质，这些应用程序涵盖移动系统和网络、计算机视觉、编程语言和机器学习。作为该项目的一部分，在定期维护的公共网站上发布。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Collage: Seamless Integration of Deep Learning Backends with Automatic Placement

Collage：深度学习后端与自动放置的无缝集成

• DOI: 10.1145/3559009.3569651
• 发表时间: 2021-11-01
• 期刊: Proceedings of the International Conference on Parallel Architectures and Compilation Techniques
• 作者: Byungsoo Jeon;Sunghyun Park;Peiyuan Liao;Sheng Xu;Tianqi Chen;Zhihao Jia
• 通讯作者: Zhihao Jia

Gemel: Model Merging for Memory-Efficient, Real-Time Video Analytics at the Edge

Gemel：模型合并可在边缘进行内存高效、实时视频分析

• 发表时间: 2023-04
• 期刊: 20th USENIX Symposium on Networked Systems Design and Implementation
• 作者: Padmanabhan, Arthi;Agarwal, Neil;Iyer, Anand;Ananthanarayan, Ganesh;Shu, Yuanchao;Karianakis, Nikolaos;Xu, Harry;Netravali, Ravi
• 通讯作者: Netravali, Ravi

TOD: GPU-Accelerated Outlier Detection via Tensor Operations

TOD：通过张量运算进行 GPU 加速异常值检测

• DOI: 10.14778/3570690.3570703
• 发表时间: 2022-11
• 期刊: Proceedings of the VLDB Endowment
• 影响因子: 2.5
• 作者: Zhao, Yue;Chen, George H.;Jia, Zhihao
• 通讯作者: Jia, Zhihao

RECL: Responsive Resource-Efficient Continuous Learning for Video Analytics

RECL：视频分析的响应式资源高效持续学习

• 发表时间: 2023-04
• 期刊: 20th USENIX Symposium on Networked Systems Design and Implementation
• 作者: Khani, Mehrdad;Ananthanarayan, Ganesh;Hsieh, Kevin;Jiang, Junchen;Netravali, Ravi;Shu, Yuanchao;Alizadeh, Mohammad;Bahl, Victor
• 通讯作者: Bahl, Victor