Collaborative Research: Highly Principled Data Science for Multi-Domain Astronomical Measurements and Analysis

合作研究：用于多领域天文测量和分析的高度原理性数据科学

• 批准号: 1811308
• 负责人: Xiao-Li Meng
• 金额: $ 18万
• 依托单位: Harvard University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-07-15 至 2022-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1811308&HistoricalAwards=false
• 关键词: Collaborative; Research; Highly; Principled; Data

Massive data resources are coming online in every conceivable area of human exploration, and particularly in fields that are heavily observation-based such as astronomy and astrophysics. To extract the most information from these data, scientists and statisticians need to conduct highly principled data science, by using methods that are scientifically justified, statistically principled, and computationally efficient. This project outlines plans to achieve this goal while addressing four specific challenges in astronomical data involving space, time and energy. The proposed research has the dual impact of more reliable statistical methods in astronomy and of new general statistical inference and computational methods. In addition to providing methods and free software, the investigators also plan to communicate to the astronomical community the benefit of principled statistical methods through workshops and sessions at conferences. A fundamental impact of the proposed research is the more general acceptance and use of principled methods among astronomers. The general methods for efficient modeling of scientific phenomena, science-driven classification and clustering, and for statistical computing, can also help to solve complex data challenges throughout the natural, social, medical, and engineering sciences.Striking advances in both space-based and terrestrial instrumentation continuously increase the quality and quantity of data available to astronomers. Observations are made across the electromagnetic spectrum and compiled into enormous catalogs of high-resolution, but heterogeneous spectrograph, imaging, and time series data. The proposed research aims to use such multi-domain astronomical measurements to better understand the physical environment, structure, and evolution of astronomical individual sources, clusters, and ultimately of the entire universe. There are four major projects. (1) The PIs will develop methodology to solve the instrument calibration problem, which is a fundamental challenge in astrophysics, by fitting scientifically motivated statistical models to data from multiple astronomical objects observed by multiple instruments. (2) The PIs propose a statistically and computationally efficient algorithm to detect the boundaries of a power law distribution prevalent in various areas of astronomy and of far-reaching importance. (3) The PIs will extend image-processing algorithms designed for detecting point sources to complex extended multi-scale structures via a post-hoc analysis, which makes the computation efficient. (4) With astronomical images exhibiting complex structure, the PIs propose to explore image segmentation methods to distinguish overlapping point sources; the algorithm achieves the flux-conserving property, which is crucial for giving physically meaningful estimates that existing methods lack. These projects all involve significant challenges in developing efficient statistical methods, designing fast computational algorithms, and balancing subtle trade-offs between complexity and practicality. With their extensive and successful track record, the PIs will address these challenges by developing inferential and efficient computational methods under highly-structured models that involve multi-scale structure and/or multiple levels of latent variables. The central theme of the proposed research is the integration and pursuit of three desiderata in each of its four projects: scientific justification, statistical principles, and computational efficiency. This triple-goal advances the development of specifically designed methods that leverage computationally efficient and statistically principled data-driven techniques which explicitly incorporate scientific understanding of the astronomical sources. This ensures that the statistical analyses enhance the scientists' ability to answer specific questions about the underlying astronomical and physical processes. This strategy requires state-of-the-art statistical inference, sophisticated scientific computing, and careful model-checking procedures, all of which have been the hallmark of the work by this team of investigators.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.

在人类探索的每一个可以想象的领域，特别是在天文学和天体物理学等大量基于观测的领域，大量的数据资源正在上线。为了从这些数据中提取最多的信息，科学家和统计学家需要通过使用科学合理、统计原则和计算效率高的方法来进行高度原则性的数据科学。该项目概述了实现这一目标的计划，同时解决涉及空间，时间和能源的天文数据的四个具体挑战。 拟议的研究具有双重影响，更可靠的统计方法在天文学和新的一般统计推断和计算方法。除了提供方法和免费软件外，研究人员还计划通过研讨会和会议向天文学界传达原则性统计方法的好处。拟议的研究的一个根本影响是天文学家更普遍地接受和使用原则性方法。科学现象的有效建模、科学驱动的分类和聚类以及统计计算的一般方法也有助于解决自然、社会、医学和工程科学中的复杂数据挑战。天基和地面仪器的惊人进步不断提高天文学家可用数据的质量和数量。观测是在整个电磁波谱范围内进行的，并被编译成大量的高分辨率但异构的光谱仪，成像和时间序列数据目录。拟议的研究旨在使用这种多域天文测量来更好地了解天文学单个源，集群以及最终整个宇宙的物理环境，结构和演化。有四个主要项目。 (1)PI将开发方法来解决仪器校准问题，这是天体物理学的一个基本挑战，通过将科学动机的统计模型拟合到多个仪器观测到的多个天文物体的数据。(2)PI提出了一种统计和计算效率高的算法，以检测在天文学的各个领域普遍存在的幂律分布的边界，具有深远的重要性。(3)PI将通过事后分析将设计用于检测点源的图像处理算法扩展到复杂的扩展多尺度结构，这使得计算高效。(4)随着天文图像表现出复杂的结构，PI建议探索图像分割方法来区分重叠的点源;该算法实现了通量守恒属性，这对于现有方法缺乏的物理意义的估计至关重要。这些项目都涉及开发有效的统计方法，设计快速计算算法以及平衡复杂性和实用性之间的微妙权衡方面的重大挑战。凭借其广泛和成功的记录，PI将通过在涉及多尺度结构和/或多层次潜变量的高度结构化模型下开发推理和高效的计算方法来应对这些挑战。拟议的研究的中心主题是在其四个项目中的每一个项目中整合和追求三个必要条件：科学论证，统计原则和计算效率。这三重目标推动了专门设计的方法的发展，这些方法利用了计算效率和统计原则的数据驱动技术，这些技术明确地结合了对天文学来源的科学理解。 这确保了统计分析提高了科学家回答有关基本天文和物理过程的具体问题的能力。这一战略需要最先进的统计推断、复杂的科学计算和仔细的模型检查程序，所有这些都是该研究团队工作的标志。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

Warp Bridge Sampling: The Next Generation

• DOI: 10.1080/01621459.2020.1825447
• 发表时间: 2016-09
• 期刊: Journal of the American Statistical Association
• 影响因子: 3.7
• 作者: Lazhi Wang;David E. Jones;X. Meng

Multiple Improvements of Multiple Imputation Likelihood Ratio Tests

多重插补似然比检验的多重改进

• DOI: 10.5705/ss.202019.0314
• 发表时间: 2022
• 期刊: Statistica Sinica
• 影响因子: 1.4
• 作者: Chan, Kin Wai;Meng, Xiao-Li
• 通讯作者: Meng, Xiao-Li

Conducting highly principled data science: A statistician’s job and joy

进行高度原则性的数据科学：统计学家的工作和乐趣

• DOI: 10.1016/j.spl.2018.02.053
• 发表时间: 2018
• 期刊: Statistics & Probability Letters
• 影响因子: 0.8
• 作者: Meng, Xiao-Li

Calibration Concordance for Astronomical Instruments via Multiplicative Shrinkage

通过乘法收缩对天文仪器进行校准一致性

• DOI: 10.1080/01621459.2018.1528978
• 发表时间: 2018
• 期刊: Journal of the American Statistical Association
• 影响因子: 3.7
• 作者: Chen, Yang;Meng, Xiao-Li;Wang, Xufei;van Dyk, David A.;Marshall, Herman L.;Kashyap, Vinay L.
• 通讯作者: Kashyap, Vinay L.