Collaborative Research: Elements: Enabling Particle and Nuclear Physics Discoveries with Neural Deconvolution

合作研究：元素：通过神经反卷积实现粒子和核物理发现

• 批准号: 2311667
• 负责人: Miguel Arratia
• 金额: $ 23.31万
• 依托单位: University of California-Riverside
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-09-01 至 2026-08-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2311667&HistoricalAwards=false
• 关键词: Collaborative; Research; Elements; Enabling; Particle

Even though large physics experiments are able to detect complex properties of reaction rates, such foundational scientific quantities are always compressed due to limitations in statistical methods and exchange platforms. The current paradigm introduces significant barriers for scientific discovery and data reusability. Comparisons between experiments with different compression schemes is challenging. Furthermore, the compression schemes necessarily throw out potentially useful information, which may be needed to explore interesting phenomena. This information loss likewise limits the long-term utility of archived data, which may be of scientific interest long after the experiment that generated it ends. Recent advancements in machine-learning methods initiated by the PIs and others solve these issues by enabling measurements directly in the un-compressed (or minimally compressed) data. However, there is currently no standard or platform for sharing such data, and therefore, no measurements of this kind with actual data have been published to date. This project builds open source cyberinfrastructure for publishing and reusing un- or minimally compressed measurements for research and educational purposes. These tools are widely applicable across physics domains and data from electron-proton collisions are used to test and benchmark the frameworks. This project serves the national interest, as stated by NSF's mission, by promoting the progress of science. The publication of minimally processed data greatly extends the practical lifetime of experimental facilities, enabling high-quality scientific analyses well beyond the time a detector is running and the researchers who collected the data. Many analyses are simply not possible with existing protocols where only limited numerical results are published alongside academic papers. Minimally compressed data can be studied without computationally expensive and often proprietary detector simulations, and are therefore of great interest for a first exposure to research by early career scientists in training.This project builds upon recent advances by the PIs and others in the development of machine machine learning solutions to measurements of reaction rates in large physics experiments. Cyberinfrastructure is created for publishing and reusing measurements created by these machine learning algorithms. The project develops an exchange format for sharing machine learning-based measurements whose data representation is neural networks, unlike the tabular, often histogram, format of traditional measurements. This format is integrated with a software platform that enables these data to be readily findable, accessible, interoperable, and reusable (FAIR). This cyberinfrastructure is tested with a prototype science pipeline, starting from a first unbinned measurement as input into the platform and ending with an analysis that reinterprets it. In the process, practical software is developed and made available to other researchers to carry out unbinned measurements and to reuse published data. These tools are integrated with widely-used frameworks in particle, nuclear, and astrophysics in order to accelerate their adaptation. Such developments are also used to broaden participation in fundamental physics and applied machine learning for undergraduate researchers, including those who would not normally have access to large experimental and computing resources. This is enabled in part by significantly reduced computational resources required to carry out forefront analysis since computationally expensive experimental tools including detector simulations are not needed.Undergraduate researchers are involved in the development and testing of the newinfrastructure.This award by the Office of Advanced Cyberinfrastructure is jointly supported by the Physics at the Information Frontier program in the Division of Physics within the Directorate for Mathematical and Physical Sciences.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.

尽管大型物理实验能够检测反应速率的复杂特性，但由于统计方法和交换平台的限制，这些基础科学量总是被压缩。 当前的范式给科学发现和数据可重用性带来了重大障碍。不同压缩方案的实验之间的比较具有挑战性。此外，压缩方案必然会丢弃潜在有用的信息，这些信息可能是探索有趣现象所需要的。这种信息丢失同样限制了存档数据的长期效用，而在生成这些数据的实验结束很久之后，这些数据可能仍具有科学意义。 PI 和其他人发起的机器学习方法的最新进展通过直接在未压缩（或最低限度压缩）数据中进行测量来解决这些问题。然而，目前还没有共享此类数据的标准或平台，因此迄今为止还没有发布此类实际数据的测量结果。 该项目构建开源网络基础设施，用于发布和重用未压缩或最低限度压缩的测量结果以用于研究和教育目的。 这些工具广泛适用于物理领​​域，电子-质子碰撞的数据用于测试和基准测试框架。 正如 NSF 的使命所述，该项目通过促进科学进步来服务于国家利益。最低限度处理的数据的发布极大地延长了实验设施的实际寿命，使高质量的科学分析远远超出了探测器运行和收集数据的研究人员的时间。现有协议根本不可能进行许多分析，因为现有协议仅与学术论文一起发表有限的数值结果。最小压缩数据可以在不需要昂贵的计算且通常是专有的探测器模拟的情况下进行研究，因此对于早期职业科学家在培训中首次接触研究非常感兴趣。该项目建立在 PI 和其他人在机器开发方面的最新进展的基础上用于测量大型物理实验中反应速率的机器学习解决方案。网络基础设施是为了发布和重用这些机器学习算法创建的测量而创建的。该项目开发了一种用于共享基于机器学习的测量的交换格式，其数据表示是神经网络，这与传统测量的表格（通常是直方图）格式不同。该格式与软件平台集成，使这些数据易于查找、访问、互操作和可重用 (FAIR)。该网络基础设施通过原型科学管道进行了测试，从第一个未分类的测量作为平台的输入开始，到重新解释它的分析结束。在此过程中，开发了实用软件并将其提供给其他研究人员进行分箱测量并重复使用已发布的数据。 这些工具与粒子、核和天体物理学中广泛使用的框架集成，以加速它们的适应。 这些进展还用于扩大本科生研究人员对基础物理和应用机器学习的参与，包括那些通常无法获得大量实验和计算资源的人。 这在一定程度上是通过显着减少进行前沿分析所需的计算资源来实现的，因为不需要包括探测器模拟在内的计算昂贵的实验工具。本科生研究人员参与了新基础设施的开发和测试。高级网络基础设施办公室颁发的这一奖项是由数学和物理科学理事会物理司的信息前沿物理项目联合支持。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进行评估，被认为值得支持 标准。