Collaborative Research: DeepCMB: New Measurements of the Cosmic Microwave Background with Deep Learning

研究宇宙微波背景（CMB）使我们能够测试早期宇宙的膨胀模型、宇宙时间内大规模结构的形成以及粒子物理学的标准模型。虽然下一代望远镜将以更高的精度观测CMB，但主要的挑战将是减少系统的不确定性。该项目将开发新的分析技术，可以从即将到来的高灵敏度CMB实验中获益。芝加哥地区现有的合作伙伴关系将被用来培训来自计算和数据科学领域代表性不足背景的学生。指导宇宙学和数据科学领域的博士后研究人员和研究生，将用于为研究环境中代表性不足的群体开发机会和技能。这项工作包括创建深度神经网络，以执行高信噪比的信息提取，以提高对r（张量与标量比）的限制，并增加在高红移和低质量下检测到的星系团的数量。该研究将使用模拟数据和南极望远镜（SPT）的数据：1)为模拟CMB数据集的快速模拟生成一个可扩展的框架；2)利用神经网络对银河系和星系外前景进行清理和消噪；3)利用深度学习分类和回归来补充现有的星系团查找算法。为这个CMB应用程序开发的工具将对其他科学以及深度学习科学本身产生交叉影响。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。

Studying the cosmic microwave background (CMB) allows us to test models of inflation in the early universe, the formation of large-scale structures across cosmic time, and the standard model of particle physics. Although next-generation telescopes will observe the CMB at new levels of precision, the primary challenge will be to reduce systematic uncertainties. This project will develop new analysis techniques that can reap the benefit of these forthcoming highly sensitive CMB experiments. Existing partnerships in the Chicago area will be leveraged to train students from underrepresented backgrounds in computation and data science. Mentoring postdoctoral researchers and graduate students in cosmology and data science will be used to develop opportunities and skillsets for underrepresented groups in research environments.The work involves creating deep neural networks to perform high signal-to-noise extraction of information, to enable improved limits on r, the tensor-to-scalar ratio, and to increase the number of detected galaxy clusters at higher redshifts and lower masses. The study will use both mock data and data from the South Pole Telescope (SPT) to: 1) produce an extensible framework for the fast simulation of mock CMB data sets; 2) use neural networks to perform galactic and extragalactic foreground cleaning and delensing; 3) use deep learning classfication and regression to complement existing galaxy clusterfinding algorithms. Tools developed for this CMB application will have cross-cutting effects on other sciences, and on the science of deep learning itself.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.