Revealing the Physical Drivers of Morphological Evolution with AI/Machine Learning and Rubin Observatory

未来十年天文观测的特点将是其数据量和复杂性的极大增加。未来几年的天文发现将依赖于社区快速处理、分析和理解海量信息的能力。在这个项目中，加州大学圣克鲁斯分校的科学家们将应用一种名为Morpheus的人工智能/机器学习(AI/ML)模型来分析和分类大规模公共天文成像测量中的天文物体。通过将人工智能/最大似然方法应用于天文数据，有可能实现原本在计算上难以处理的分析。通过发布Morpheus作为一个开放的框架，供其他科学家应用于他们自己的数据集，这项研究将使概率形态成为广泛的银河系外成像测量的可行测量，从而大大增加对星系形成的知识。该项目的目标包括1)进一步了解形态和支配星系形成的物理之间的联系，以及2)降低将强大的AI//ML方法应用于天文数据集的门槛。作为该项目的一部分，该团队还将为研究生和博士后建立每年一次的免费研讨会，以开发高质量、可转移和可扩展的专业网站。这些活动将提高年轻研究人员在天文学和天体物理学领域的能见度，同时为他们提供持久的在线足迹，以展示他们的专业活动。拟议的研究将应用用于天文数据分析的Morpheus深度学习框架，对大规模公共调查数据中的星系进行语义分割、源提取和形态分类。Morpheus框架利用AI/ML技术提供逐个像素的图像分类，检测对象，生成相应的分割图，然后量化每个像素属于一类天文对象的模型概率。该团队将把Morpheus纳入Rubin Science平台，使用Rubin和基于空间的数据的组合进行验证，并将其应用于最初的LSST数据发布。该团队还将使用得到的Morpheus形态来研究形态和其他星系属性之间的相关性。该项目还支持加州大学圣克鲁斯分校的鲁宾天文台科学验证活动。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。

The hallmark of astronomical surveys over the next decade will be their vastly increased data volumes and complexity. Astronomical discoveries in coming years will rely on the ability of the community to rapidly process, analyze, and understand enormous amounts of information. In this project, scientists at the University of California, Santa Cruz, will apply an Artificial Intelligence/Machine Learning (AI/ML) model called Morpheus to analyze and classify astronomical objects in large-scale, public astronomical imaging surveys. Through the application of AI/ML methods to astronomical data it is possible to enable analyses that are otherwise computationally intractable. By releasing Morpheus as an open framework for other scientists to apply on their own datasets, this research will substantially augment the knowledge of galaxy formation by making probabilistic morphology a feasible measurement for a wide range of extragalactic imaging surveys. Goals of the project include 1) furthering the understanding of the connection between morphology and the physics that govern galaxy formation and 2) lowering the bar for the application of powerful AI//ML methodologies to astronomical datasets. As part of this project, the team will also establish a yearly free workshop for graduate students and postdocs to develop high-quality, transferrable, and extendable professional websites. These activities will increase the visibility of young researchers in astronomy and astrophysics, while providing them with a durable on-line footprint for featuring their professional activities.The proposed research will apply the Morpheus deep learning framework for astronomical data analysis to perform semantic segmentation, source extraction, and morphological classification of galaxies in large scale public survey data. The Morpheus framework leverages AI/ML technology to provide pixel-by-pixel classifications of images, detecting objects, producing corresponding segmentation maps, and then quantifying the model probability that each pixel belongs a class of astronomical object. The team will incorporate Morpheus into the Rubin Science Platform, validate it with a combination of Rubin and space-based data, and apply it to the initial LSST data releases. The team will also use the resulting Morpheus morphologies to investigate the correlations between morphology and other galaxy properties. This project also supports Rubin Observatory science verification activities at UC Santa Cruz.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.