Elements: Scalable Bayesian Software for Interpreting Astronomical Images

BLISS(贝叶斯光源分离器)项目是一项跨学科的研究工作，旨在开发一种软件工具，使天文学家能够利用机器学习的最新进展。通过利用这些进步，天文学家可以快速分析大量复杂的数据，以了解我们宇宙的本质。该项目还通过举办一系列讲习班，促进软件开发和机器学习方面的技术熟练程度，让更多的受众参与并教育他们。作为该项目的一部分开发的软件工具，将使天文学家能够更容易地使用贝叶斯统计方法来解释天文测量的图像数据。贝叶斯方法擅长不确定性量化和数据集成，这两种能力将在分析下一代天文测量产生的海量数据方面至关重要。更广泛地采用贝叶斯分析来解释天文图像的一个主要障碍是计算：众所周知，贝叶斯推理的计算要求很高。第二个主要障碍是社会障碍：到目前为止，新的贝叶斯方法是由统计学家单独开发的，很少被整合到天文学工作流程中，因为这两个学科的从业者都不清楚如何实现这一点。BLISS项目解决了这些计算和社区集成挑战。为了克服计算挑战，BLISS利用了贝叶斯推理方法的最新进展，包括使用深度学习、变分推理和GPU加速。为了确保即时和可持续的社区使用，BLISS的开发是由领域专家确定的需求指导的，他们自己也准备参与BLISS的开发，并热衷于将BLIS整合到他们团队的数据分析工作流程中。该项目得到了计算机和信息科学与工程局的高级网络基础设施办公室、数学科学和物理科学局的天文科学部的支持。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The BLISS (Bayesian Light Source Separator) Project is an interdisciplinary research effort to develop a software tool that allows astronomers to make use of the latest advances in machine learning. By harnessing these advances, astronomers can rapidly analyze vast quantities of complex data to understand the nature of our universe. This project also engages and educates a wider audience through a workshop series that promotes technical proficiency in software development and machine learning.The software tool, developed as part of this project, will allow astronomers to more easily access Bayesian statistical methods to interpret image data from astronomical surveys. Bayesian methods excel at uncertainty quantification and data integration, two capabilities that will be critical in analyzing the deluge of data produced by next-generation astronomical surveys. One major barrier to the more widespread adoption of Bayesian analysis for interpreting astronomical images is computational: Bayesian inference is notoriously computationally demanding. A second major barrier is social: up to now, novel Bayesian methods have been developed in isolation by statisticians and have rarely been integrated into astronomy workflows because it is unclear to practitioners in either discipline how this can be accomplished. The BLISS Project addresses both these computational and community integration challenges. To overcome the computational challenges, BLISS leverages recent advances in Bayesian inference methodology, including the use of deep learning, variational inference, and GPU acceleration. To ensure immediate and sustainable community use, development of the BLISS is guided by needs identified by domain experts, who are themselves prepared to participate in BLISS's development and are enthusiastic about integrating BLISS into their teams' data analysis workflows.This project is supported by the Office of Advanced Cyberinfrastructure in the Directorate for Computer & Information Science & Engineering, the Division of Mathematical Sciences and the Division of Astronomical Sciences in the Directorate for Mathematical and Physical Sciences.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.