Deep-Learning for Galaxy Morphology in the Big Data Era

天文学正在进入大数据时代。 不久将从大规模调查中获得的大量数据对于理解星系演化将是无价的。 然而，从极其丰富的数据集中提取和解释信息是一项挑战，迄今为止还没有足够有效的解决方案。 这是一个定制深度学习算法的项目，该算法已成功用于模式识别重要的其他领域，以快速准确地测量星系形态。 长期目标是开发将大数据转化为天体物理学大发现的算法。 所有的算法和分类都将提供给更普遍的使用。 首席研究员致力于支持来自代表性不足的群体的科学家，通过招聘实践，在美国和其他国家教学，并帮助发展中国家建立产生尖端科学的能力。形态学是限制星系形成模型的关键可观察因素，但量化形态学目前是一个耗时的过程，受到大数据过渡的严重影响。 深度学习算法可能是答案。 在第一阶段，算法将从斯隆数字巡天（SDSS）提供的大量形态分类中学习。 下一阶段研究如何将从SDSS获得的知识转移到暗能量调查（DES）的图像分析中。 还将研究使用模拟图像来加速学习过程。 这是对星系结构的其他方面进行自动分类的第一步。 长期目标是开发将大数据转化为天体物理学大发现的算法。 使用深度学习衍生的形态将说明数据到发现的过程。 向来访的天文学家、教育工作者和硕士生教授使用SDSS和DES数据库的专业知识，将有助于确保对建立这些数据库的投资产生更广泛的全球影响。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Astronomy is entering the Big Data era. The wealth of data which will soon be available from massive surveys will be invaluable for understanding galaxy evolution. However, extracting and interpreting information from enormously rich datasets is a challenge with no sufficiently efficient, demonstrated solutions to date. This is a project to tailor Deep Learning algorithms, which have been used successfully in other fields where pattern recognition matters, to measure galaxy morphologies quickly and accurately. The long-term goal is to develop algorithms which transform Big Data into Big Discovery in astrophysics. All algorithms and classifications will be made available for more general use. The principal researcher is committed to supporting scientists from under-represented groups, through hiring practices, teaching in the US and in other countries, and by helping to build the capacity for developing countries to produce cutting-edge science.Morphology is a key observable for constraining galaxy formation models, but quantifying morphology is currently a time-consuming process, severely compromised by the big-data transition. Deep Learning algorithms may be the answer. In the first phase, algorithms will learn from the large set of morphological classifications available from the Sloan Digital Sky Survey (SDSS). The next phase studies how to transfer the knowledge gained from SDSS into analysis of images from the Dark Energy Survey (DES). The use of simulated images to accelerate the learning process will also be studied. This is the first step towards the automated classification of other aspects of galaxy structure. The long-term goal is to develop algorithms which transform Big Data into Big Discovery in astrophysics. Work with Deep Learning derived morphologies will illustrate the data-to-discovery process. Teaching expertise in using the SDSS and DES databases to visiting astronomers, educators and masters-level students will help to ensure a wider global impact of the investment in building these databases.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.