Collaborative Research: CDS&E: Optimizing discovery with multi-epoch photometric survey data

当前和即将到来的多历元数字巡天正在以前所未有的广度和灵敏度探测时域，产生大量关于可变和瞬变物体的信息，从而彻底改变天文学。该项目将开发新的统计和机器学习方法和软件，以利用多历元成像调查数据改进天体物理对象的检测、表征和分类。这种科学技术将在各种地面和天基调查环境中发挥作用。新的分类器还将有助于优化有限的后续资源的分配。所有算法都将在公开的开源软件中实现。这个团队包括天文学家、应用数学家和统计学家，这个项目需要跨学科的创新，并将加强他们的伙伴关系。这项工作将培训一名应用数学研究生，以解决天文学中的数据科学问题，并包括支持在高级统计和机器学习方法方面培训不同的年轻天文学家。目标检测、表征和分类是从调查数据中提取科学的基础。该项目的主要工作是开发基于多波段光曲线形状的可变和瞬态物体分类的新算法。第二种方法将使用贝叶斯神经网络，使用改进的先验和改进的解空间探索。配套组件解决了源似然函数的最佳规范和使用，开发了源不确定性的更准确表示，致力于对昏暗、可变物体的最佳检测，并对宇宙种群进行人口统计学建模。该奖项反映了美国国家科学基金会的法定使命，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。

Current and forthcoming multi-epoch digital sky surveys are revolutionizing astronomy by probing the time domain with unprecedented breadth and sensitivity, producing vast amounts of information on variable and transient objects. This project will develop new statistical and machine learning methods and software to improve the detection, characterization, and classification of astrophysical objects using multi-epoch imaging survey data. This science-enabling technology will be useful in diverse ground- and space-based survey settings. New classifiers will also help to optimize the allocation of limited follow-up resources. All algorithms will be implemented in public, open-source software. The team includes astronomers, applied mathematicians, and statisticians, and this project requires innovation across disciplines and will enhance their partnership. The work will train a graduate student in applied math to work on data science problems in astronomy, and includes support to train a diverse population of young astronomers in advanced statistics and machine learning methods.Object detection, characterization, and classification are fundamental to extracting science from survey data. The main work of this project is to develop new algorithms for classifying variable and transient objects based on the shapes of their multiband light curves. A second approach will use Bayesian neural networks, using improved priors and improved solution space exploration. Companion components address optimal specification and use of source likelihood functions, developing more accurate representations of source uncertainty, working towards the optimal detection of dim, variable objects, and doing demographic modeling of cosmic populations.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.