Rubin LSST 暗能量科学合作的最新 AI/ML 机遇
天体物理仪器与方法
2026-01-22 v1 宇宙学与河外天体物理
人工智能
机器学习
机器学习
摘要
Vera C. Rubin Observatory 的 Legacy Survey of Space and Time (LSST) 将产生前所未有的异构天文数据 (图像、目录和警报), 挑战传统的数据分析管道. 暗能量科学合作 (DESC) 旨在从这些数据中推导出对暗能量和暗物质的稳健约束, 要求方法具备统计效力、可扩展性和操作可靠性. 人工智能和机器学习 (AI/ML) 已经嵌入 DESC 科学工作流程中, 包括光度红移、暂态分类、弱引力推断和宇宙模拟等. 然而, 其在精密宇宙学中的实用性取决于可信赖的不确定性量化、对协变量漂移和模型误指定的鲁棒性, 以及在科学管道中的可重复集成. 本白皮书调查了 DESC 主要宇宙探针和交叉分析中 AI/ML 的当前格局, 揭示了相同核心方法和基本挑战在不同科学案例中反复出现. 由于解决这些交叉挑战的进展将同时惠及多个探针, 我们确定了关键方法学研究优先事项, 包括大规模贝叶斯推断、物理信息方法、验证框架以及发现主动学习. 同时, 我们也探讨了最新基础模型方法和 LLM 驱动的智能体 AI 系统在 DESC 工作流程中的潜在影响, 前提是其部署伴随严格的评估和治理. 最后, 我们讨论了成功部署这些新方法所需的关键软件、计算、数据基础设施和人力资本要求, 并考虑与外部参与者的更广泛协调所带来的风险与机遇.
引用
@article{arxiv.2601.14235,
title = {Opportunities in AI/ML for the Rubin LSST Dark Energy Science Collaboration},
author = {LSST Dark Energy Science Collaboration and Eric Aubourg and Camille Avestruz and Matthew R. Becker and Biswajit Biswas and Rahul Biswas and Boris Bolliet and Adam S. Bolton and Clecio R. Bom and Raphaël Bonnet-Guerrini and Alexandre Boucaud and Jean-Eric Campagne and Chihway Chang and Aleksandra Ćiprijanović and Johann Cohen-Tanugi and Michael W. Coughlin and John Franklin Crenshaw and Juan C. Cuevas-Tello and Juan de Vicente and Seth W. Digel and Steven Dillmann and Mariano Javier de León Dominguez Romero and Alex Drlica-Wagner and Sydney Erickson and Alexander T. Gagliano and Christos Georgiou and Aritra Ghosh and Matthew Grayling and Kirill A. Grishin and Alan Heavens and Lindsay R. House and Mustapha Ishak and Wassim Kabalan and Arun Kannawadi and François Lanusse and C. Danielle Leonard and Pierre-François Léget and Michelle Lochner and Yao-Yuan Mao and Peter Melchior and Grant Merz and Martin Millon and Anais Möller and Gautham Narayan and Yuuki Omori and Hiranya Peiris and Laurence Perreault-Levasseur and Andrés A. Plazas Malagón and Nesar Ramachandra and Benjamin Remy and Cécile Roucelle and Jaime Ruiz-Zapatero and Stefan Schuldt and Ignacio Sevilla-Noarbe and Ved G. Shah and Tjitske Starkenburg and Stephen Thorp and Laura Toribio San Cipriano and Tilman Tröster and Roberto Trotta and Padma Venkatraman and Amanda Wasserman and Tim White and Justine Zeghal and Tianqing Zhang and Yuanyuan Zhang},
journal= {arXiv preprint arXiv:2601.14235},
year = {2026}
}
备注
84 pages. This is v1.0 of the DESC's white paper on AI/ML, a collaboration document that is being made public but which is not planned for submission to a journal