Eyes on the future: optimizing science output for next generation surveys with joint crowdsourced and automated classification techniques

未来的星系调查将产生如此多的数据，天文学家将不再能够依靠他们以前的分类方法，以便通过宇宙历史提取星系形成和演化的科学。虽然“众包”的星系分类利用了大量志愿者的劳动资源，但即使是像银河动物园（GZ）这样的大型努力也无法跟上。该项目将以GZ数据库为基础，将方法扩展到宇宙的其他时代，同时将公民科学工作的结果与机器学习结合起来开发新的自动分类工具。这项研究利用公众和更广泛社区参与的研究价值，并采用新一代智能计算机方法，将两者的优点结合起来。虽然众包星系分类已经在斯隆数字巡天（SDSS） 10年的数据中证明了它们的价值，但要使它们成为下一代巡天数据处理管道的标准组成部分，仍然存在两个主要挑战。第一个是证明该方法在高红移的实用性，在高红移中，更多的星系具有不规则或块状的形态。第二种观点承认，即使是众包也没有能力处理未来的数据量和速率，这需要新的更复杂的机器分类算法。该项目将为高红移众包数据和模拟星系开发目录，并开发一种新的自动化分类工具，该工具可扩展到更高的红移，并具有适应多个星系调查的训练管道。它包括三个科学项目：(a)直接约束星系的大小和质量增长率；(b)测量条形盘与活动星系核的燃料之间的任何关系；(c)量化盘子结构的人口统计学和演化。使用参数和非参数技术自动化形态分类的努力在SDSS中已经相当成功，但没有扩展到导出必要的详细结构参数。广州众包项目取得了超乎预期的成功，从志愿者工作中提供了科学上可行的参数，并发表了一百多篇同行评议的论文。这项研究将扩展这两种方法，为未来更大规模的调查做准备，这些调查预计每晚产生与十年SDSS一样多的数据。将要编制的高级目录和将要使用的新分类算法将向公众发布，这将是社区的宝贵资源。与非正式的指导性探究项目一起，GZ将被实施到本科天文学课程中。参与这项工作的学生将获得明确的博士项目，并获得对他们未来职业生涯有价值的技术技能。

Future galaxy surveys will produce so much data that astronomers will no longer be able to rely on their previous methods of classifying them in order to extract the science of galaxy formation and evolution through cosmic history. Although "crowd-sourced" galaxy classifications tap into a vast resource of volunteer labor, even major efforts like the Galaxy Zoo (GZ) will not be able to keep up. This project will build on the GZ database, extending methods to other epochs in the Universe, and simultaneously use the results of the citizen science work together with machine learning to develop new automated classification tools. Using the proven research value of involving the public and the broader community, and with a new generation of intelligent computer methods, this study will build on the best of both.While crowd-sourced galaxy classifications have proven their worth on a decade of data from the Sloan Digital Sky Survey (SDSS), there remain two major challenges to making them a standard component of the data processing pipelines for the next generation of surveys. The first is proving the utility of the method at high redshifts, where more galaxies have irregular or clumpy morphologies. The second acknowledges that even crowdsourcing does not have the capacity for the data volume and rates that are to come, requiring new more sophisticated machine classification algorithms. This project will develop catalogs for high-redshift crowd-sourced data and for simulated galaxies, and develop a new automated classification tool that extends to higher redshifts with a training pipeline adaptable to multiple galaxy surveys. It includes three science projects: (a) directly constrain galaxy size and mass growth rates; (b) measure any relationship between bar-dominated disks and fueling of active galactic nuclei, and (c) quantify the demographics and evolution of disk sub-structures. Efforts to automate morphological classifications using parametric and non-parametric techniques have been reasonably successful for SDSS, but did not extend to deriving the necessary detailed structural parameters. The GZ crowd-sourcing project has been successful beyond expectations, providing scientifically viable parameters from volunteer work and leading to over a hundred peer-reviewed papers. This study will extend both of these approaches in preparation for much larger future surveys, which expect to produce as much data per night as ten years of SDSS.The high-level catalogs to be produced, and the new classification algorithms to be used, are to be released publically, and will be a valuable resource for the community. Along with informal guided-inquiry projects, GZ will be implemented into undergraduate astronomy courses. Students involved in this work will get well-defined PhD projects and acquire technical skills valuable in their future professional careers.