Collaborative Research: CNS Core: Medium: Dynamic Data-driven Systems - Theory and Applications

合作研究：CNS 核心：媒介：动态数据驱动系统 - 理论与应用

• 批准号: 2106299
• 负责人: Mohammadhassan Hajiesmaili
• 金额: $ 36万
• 依托单位: University of Massachusetts Amherst
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2106299&HistoricalAwards=false
• 关键词: Collaborative; Research; CNS; Core; Medium

Modern computer systems must be continually optimized in a data-driven manner to maintain performance, even as their deployment and workload environments change. This holds for traditional systems like content delivery networks and emerging architectures such as edge/cloud systems. The design of dynamic data-driven systems requires both theoretical advancements and new systems architectures. A key challenge is a tradeoff between optimality, i.e., choosing an optimal deployment for the current environment in terms of performance and/or cost, and smoothness, i.e., ensuring that the deployment changes are not too costly at any point. This project seeks to develop tools at the intersection of machine learning and optimization that enable systems to balance between optimality and smoothness. Further, this project deploys and empirically evaluates these tools in the context of 360 video streaming as a representative case study. Smoothness is not a traditional system performance measure, and so it is typically enforced only in ad hoc ways by existing systems. However, it is a crucial consideration for systems that seek to continuously optimize their configuration since the switching costs associated with changing configurations can be significant. Managing the tradeoff between optimality and smoothness in a rigorous fashion can lead to dramatic improvements; however, it is challenging since it requires a robust data-driven design that can determine whether it is worth incurring a switching cost in the present, without knowledge of the future environment. This project develops analytic tools that enable the design of algorithms for dynamic systems that balance optimality and smoothness through the integration of data-driven and optimization approaches. There are also planned test-bed deployment activities for 360 video streaming. The project will provide new foundational tools for the design of dynamic systems across multiple application areas. While we choose video streaming as our target application, the proposed fundamental research is applicable much more broadly. Notably, this project broadens the participation of underrepresented groups in STEM areas through programs at both K-12 and undergraduate levels. Planned activities include developing accelerated mathematics programs for middle-school students, summer programs for middle-school and high-school students, and summer research programs for undergraduate students. This is a collaborative project with investigators from the University of Massachusetts Amherst, California Institute of Technology, and the State University of New York at Stony Brook. The results of this project will be maintained on the project website at https://groups.cs.umass.edu/hajiesmaili/soco/. These will include technical reports of the research findings, software prototypes of the algorithms designed, datasets, and experimental results collected for the 360 video streaming experiments.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.

现代计算机系统必须以数据驱动的方式持续优化，以保持性能，即使它们的部署和工作负载环境发生变化。这适用于内容交付网络等传统系统和边缘/云系统等新兴架构。动态数据驱动系统的设计既需要理论的进步，也需要新的系统架构。一项关键挑战是在最优化(即在性能和/或成本方面为当前环境选择最佳部署)和平稳性(即确保部署变化在任何时候都不会太昂贵)之间进行权衡。该项目寻求在机器学习和优化的交叉点上开发工具，使系统能够在最优性和流畅性之间取得平衡。此外，本项目在360视频流的背景下部署并经验性地评估了这些工具，作为典型的案例研究。流畅性不是传统的系统性能衡量标准，因此它通常仅由现有系统以特殊方式强制执行。但是，对于寻求持续优化其配置的系统来说，这是一个重要的考虑因素，因为与更改配置相关的切换成本可能会很高。以严格的方式管理最优性和流畅性之间的权衡可以带来显著的改进；然而，这是具有挑战性的，因为它需要强大的数据驱动设计，该设计可以在不了解未来环境的情况下确定目前是否值得承担切换成本。该项目开发的分析工具能够为动态系统设计算法，通过整合数据驱动和优化方法来平衡最佳性和平稳性。此外，还计划进行360视频流的试验台部署活动。该项目将为跨多个应用领域的动态系统设计提供新的基础工具。虽然我们选择视频流作为我们的目标应用，但提出的基础研究适用范围更广。值得注意的是，该项目通过K-12和本科两个级别的课程，扩大了STEM领域中代表性不足的群体的参与。计划的活动包括为中学生开发加速数学课程，为初中生和高中生开发暑期课程，为本科生开发暑期研究课程。这是一个与马萨诸塞大学阿默斯特分校、加州理工学院和纽约州立大学石溪分校的研究人员合作的项目。该项目的结果将在项目网站https://groups.cs.umass.edu/hajiesmaili/soco/.上维护这些将包括研究结果的技术报告，所设计的算法的软件原型，数据集，以及为360个视频流实验收集的实验结果。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势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

Contextual Combinatorial Bandits with Probabilistically Triggered Arms

• DOI: 10.48550/arxiv.2303.17110
• 发表时间: 2023-03
• 作者: Xutong Liu;Jinhang Zuo;Siwei Wang;John C.S. Lui;M. Hajiesmaili;A. Wierman;Wei Chen

Pareto-Optimal Learning-Augmented Algorithms for Online Conversion Problems

在线转换问题的帕累托最优学习增强算法

• 发表时间: 2021
• 期刊: Advances in Neural Information Processing Systems 34 (NeurIPS 2021
• 作者: Bo Sun, Russell Lee

Distributed Bandits with Heterogeneous Agents

• DOI: 10.1109/infocom48880.2022.9796901
• 发表时间: 2022-01
• 期刊: IEEE INFOCOM 2022 - IEEE Conference on Computer Communications
• 作者: Lin Yang;Y. Chen;M. Hajiesmaili;John C.S. Lui;D. Towsley

Online Peak-Aware Energy Scheduling with Untrusted Advice

• DOI: 10.1145/3447555.3464860
• 发表时间: 2021-06
• 期刊: Proceedings of the Twelfth ACM International Conference on Future Energy Systems
• 作者: Russell Lee;Jessica Maghakian;M. Hajiesmaili;Jian Li;R. Sitaraman;Zhenhua Liu

On-Demand Communication for Asynchronous Multi-Agent Bandits

• DOI: 10.48550/arxiv.2302.07446
• 发表时间: 2023-02
• 作者: Y. Chen;L. Yang;Xuchuang Wang;Xutong Liu;M. Hajiesmaili;John C.S. Lui;D. Towsley