Framework: Software: Next-Generation Cyberinfrastructure for Large-Scale Computer-Based Scientific Analysis and Discovery

框架：软件：用于大规模计算机科学分析和发现的下一代网络基础设施

• 批准号: 1835443
• 负责人: Alan Edelman
• 金额: $ 349.86万
• 依托单位: Massachusetts Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1835443&HistoricalAwards=false
• 关键词: Framework; Software; Next; Generation; Cyberinfrastructure

Recent revolutions in data availability have radically altered activities across many fields within science, industry, and government. For instance, contemporary simulations medication properties can require the computational power of entire data centers, and recent efforts in astronomy will soon generate the largest image datasets in history. In such extreme environments, the only viable path forward for scientific discovery hinges on the development and exploitation of next-generation computational cyberinfrastructure of supercomputers and software. The development of this new computational infrastructure demands significant engineering resources, so it is paramount to maximize the infrastructure's potential for high impact and wide adoption across as many technical domains as possible. Unfortunately, despite this necessity, existing development processes often produce software that is limited to specific hardware, or requires additional expertise to use properly, or is overly specialized to a specific problem domain. Such "single-use" software tools are limited in scope, leading to underutilization by the wider scientific community. In contrast, this project seeks to develop methods and software for computer-based scientific analysis that are sufficiently powerful, flexible and accessible to (i) enable domain experts to achieve significant advancements within their domains, and (ii) enable innovative use of advanced computational techniques in unexpected scientific, technological and industrial applications. This project will apply these tools to a wide variety of specific scientific challenges faced by various research teams in astronomy, medicine, and energy management. These teams plan on using the proposed work to map out new star systems, develop new life-saving medications, and design new power systems that will deliver more energy to a greater number of homes and businesses at a lower cost than existing systems. Finally, this project will seek to leave a legacy of sustained societal benefit by educating students and practitioners in the broader scientific and engineering communities via exposure to state-of-the-art computational techniques. Through close collaboration with research teams in statistical astronomy, pharmacometrics, power systems optimization, and high-performance computing, this project will deliver cyberinfrastructure that will effectively and effortlessly enable the next generation of computer-based scientific analysis and discovery. To ensure the practical applicability of the developed cyberinfrastructure, the project will focus on three target scientific applications: (i) economically viable decarbonization of electrical power networks, (ii) real-time analysis of extreme-scale astronomical image data, and (iii) pharmacometric modeling and simulation for drug analysis and discovery. While tackling these specific problems will constitute an initial stress test of the proposed cyberinfrastructure, it is the ultimate goal of the project that the developed tools be sufficiently performant, accessible, composable, flexible and adaptable to be applied to the widest possible range of problem domains. To achieve this vision, the project will build and improve various software tools for computational optimization, machine learning, parallel computing, and model-based simulation. Particular attention will be paid to the proposed cyberinfrastructure's composability with new and existing tools for scientific analysis and discovery. The pursuit of these goals will require the design and implementation of new programming language abstractions to allow close integration of high-level language features with low-level compiler optimizations. Furthermore, maximally exploiting proposed cyberinfrastructure will require research into new methods that combine state-of-the-art techniques from optimization, machine learning, and high-performance computing.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.

最近数据可用性方面的革命从根本上改变了科学、工业和政府内部许多领域的活动。例如，当代的药物特性模拟可能需要整个数据中心的计算能力，最近在天文学方面的努力将很快产生历史上最大的图像数据集。在这种极端环境中，科学发现的唯一可行途径取决于超级计算机和软件的下一代计算网络基础设施的开发和利用。这种新的计算基础设施的开发需要大量的工程资源，因此，最大限度地发挥基础设施的潜力，使其在尽可能多的技术领域产生高影响和广泛采用是至关重要的。不幸的是，尽管有这种必要性，但现有的开发过程通常会生成仅限于特定硬件的软件，或者需要额外的专业知识才能正确使用，或者过于专用于特定的问题领域。这种“一次性使用”的软件工具范围有限，导致更广泛的科学界利用不足。与此形成对比的是，该项目致力于开发以计算机为基础的科学分析方法和软件，这些方法和软件具有足够的功能、灵活性和可及性，以便(I)使领域专家能够在其领域内取得重大进展，(Ii)能够在意外的科学、技术和工业应用中创新地使用先进的计算技术。该项目将把这些工具应用于天文学、医学和能源管理领域的各个研究团队所面临的各种具体科学挑战。这些团队计划利用拟议的工作来规划新的恒星系统，开发新的救命药物，并设计新的电力系统，以比现有系统更低的成本向更多的家庭和企业提供更多的能量。最后，该项目将寻求通过接触最先进的计算技术来教育更广泛的科学和工程界的学生和从业者，从而留下可持续的社会利益的遗产。通过与统计天文学、药物计量学、电力系统优化和高性能计算方面的研究团队的密切合作，该项目将提供网络基础设施，从而有效和毫不费力地实现下一代基于计算机的科学分析和发现。为了确保已开发的网络基础设施的实际适用性，该项目将侧重于三个目标科学应用：(1)经济上可行的电力网络脱碳；(2)对极端尺度天文图像数据的实时分析；(3)药物分析和发现的药物计量建模和模拟。虽然解决这些具体问题将构成对拟议的网络基础设施的初步压力测试，但该项目的最终目标是，所开发的工具具有足够的性能、可访问性、可组合性、灵活性和适应性，以适用于尽可能广泛的问题领域。为了实现这一愿景，该项目将建立和改进各种软件工具，用于计算优化、机器学习、并行计算和基于模型的模拟。将特别注意拟议的网络基础设施与新的和现有的科学分析和发现工具的组合能力。为了实现这些目标，将需要设计和实现新的编程语言抽象，以实现高级语言功能与低级编译器优化的紧密结合。此外，最大限度地利用拟议的网络基础设施将需要研究结合来自优化、机器学习和高性能计算的最先进技术的新方法。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

Reverse-Mode Automatic Differentiation and Optimization of GPU Kernels via Enzyme

• DOI: 10.1145/3458817.3476165
• 发表时间: 2021-11
• 期刊: SC21: International Conference for High Performance Computing, Networking, Storage and Analysis
• 作者: William S. Moses;Valentin Churavy;Ludger Paehler;J. Hückelheim;S. Narayanan;Michel Schanen;J. Doerfert

Accelerating Simulation of Stiff Nonlinear Systems using Continuous-Time Echo State Networks

使用连续时间回波状态网络加速刚性非线性系统的仿真

• 发表时间: 2021
• 期刊: Proceedings of the AAAI 2021 Spring Symposium on Combining Artificial Intelligence and Machine Learning with Physical Sciences
• 作者: Anantharaman, Ranjan;Ma, Yingbo;Gowda, Shashi;Laughman, Chris;Shah, Viral;Edelman, Alan;Rackauckas, Chris
• 通讯作者: Rackauckas, Chris

Low-Rank Univariate Sum of Squares Has No Spurious Local Minima

低秩单变量平方和没有虚假局部最小值

• DOI: 10.1137/22m1516208
• 发表时间: 2023
• 期刊: SIAM Journal on Optimization
• 影响因子: 3.1
• 作者: Legat, Benoît;Yuan, Chenyang;Parrilo, Pablo
• 通讯作者: Parrilo, Pablo

Rapid software prototyping for heterogeneous and distributed platforms

• DOI: 10.1016/j.advengsoft.2019.02.002
• 发表时间: 2019-06-01
• 期刊: ADVANCES IN ENGINEERING SOFTWARE
• 影响因子: 4.8
• 作者: Besard, Tim;Churavy, Valentin;De Sutter, Bjorn
• 通讯作者: De Sutter, Bjorn

Differential methods for assessing sensitivity in biological models.

• DOI: 10.1371/journal.pcbi.1009598
• 发表时间: 2022-06
• 期刊: PLoS computational biology
• 影响因子: 4.3