BIGDATA: IA: Query-by-example for Big astronomical Data

虽然现代观测天体物理学中最重要的项目产生非常大的数据集，但计算方法落后，并且难以有效地分析这些数据。尽管当前和未来的天文调查将产生世界上最大的公共数据库，但将这些数据转化为科学发现的方法尚不存在。该项目将使用无监督机器学习技术，开发一个自动示例查询系统，根据星系与其他星系的相似性对星系进行分类。这种能力目前还不存在，它可以大大提高数字巡天的经验和发现能力。该项目有一个教育组成部分，特别侧重于代表性不足的少数民族学生的本科研究。现有的，特别是计划中的，调查可以对数十亿个星系进行成像，这使得通过计算机分析研究稀有星系的能力变得绝对必要。这些不寻常的天体对于理解关于早期、现在和未来宇宙的最基本问题至关重要，因为它们携带着关于物体相互作用、它们的形成和演化的历史的关键信息。待开发的系统将拍摄某个（通常是特殊的）星系的图像，由研究人员确定为感兴趣的星系，并将在数百万个星系中搜索，以找到视觉上与查询星系最相似的星系。因为研究不寻常的星系并对其性质做出科学结论需要一定的种群，从中得出可以与其他系统进行比较的特性，这种能力和由此产生的列表将大大增加从天空调查中发现的能力，优化这些重要而昂贵的研究仪器的科学回报。该项目与现有的努力相结合，以吸引代表性不足的少数民族，在他们本科学位的后期增加更多的高级研究培训。像这样的研究总是有机会向公众宣传，团队希望为形成他们所在地区的大数据中心做出贡献。

While the most significant projects in modern observational astrophysics generate very large data sets, the computational methodology lags behind, and has trouble effectively analyzing these data. Although current and future astronomical surveys will produce the world's largest public databases, the methods to turn these data into scientific discoveries do not yet exist. This project will develop an automatic query-by-example system for classifying galaxies by their similarities to other galaxies, using unsupervised machine learning techniques. This capability does not currently exist and it can substantially enhance the experience and discovery power of digital sky surveys. The project has an educational component, focusing in particular on undergraduate research for under-represented minority students.Existing, and especially planned, surveys can image billions of galaxies, making the ability to study rare galaxies through computer analysis absolutely essential. These uncommon objects are critical for understanding the most fundamental questions about the early, present, and future universe, as they carry crucial information on the history of the interactions of objects, their formation, and their evolution. The system to be developed will take an image of a certain (normally peculiar) galaxy, identified by the researcher as being of interest, and will search through millions of galaxies to find the visually most similar galaxies to the query galaxy. Because studying unusual galaxies and making scientific conclusions about their nature requires a certain population from which to derive properties that can be compared to other systems, this capability and the resulting listings will greatly increase the ability to make discoveries from sky surveys, optimizing the scientific return of these important and expensive research instruments.This project connects with existing efforts to attract under-represented minorities, adding more advanced research training in the later years of their undergraduate degree. Studies like this always include opportunities for public outreach, and the team expects to contribute to the forming big data hub in their area.