中文

多模态宇宙:利用 100TB 天文科学数据实现大规模机器学习

天体物理仪器与方法 2024-12-04 v1 星系天体物理 太阳与恒星天体物理

摘要

我们介绍了 MULTIMODAL UNIVERSE,一个大规模多模态天文科学数据集,专为促进机器学习研究而编译。总体而言,MULTIMODAL UNIVERSE 包含数亿天文观测数据,构成 100\,TB 的多通道和超光谱图像、光谱、多变量时间序列,以及各类相关科学测量数据和 "metadata"。此外,我们还包括一系列代表 astrophysics 中标准机器学习实践的基准任务。这一巨大的数据集将促进针对科学应用开发大型多模态模型。所有用于编译 MULTIMODAL UNIVERSE 的代码,以及数据获取方式说明,均可在 https://github.com/MultimodalUniverse/MultimodalUniverse 查阅。

关键词

引用

@article{arxiv.2412.02527,
  title  = {The Multimodal Universe: Enabling Large-Scale Machine Learning with 100TB of Astronomical Scientific Data},
  author = {The Multimodal Universe Collaboration and Jeroen Audenaert and Micah Bowles and Benjamin M. Boyd and David Chemaly and Brian Cherinka and Ioana Ciucă and Miles Cranmer and Aaron Do and Matthew Grayling and Erin E. Hayes and Tom Hehir and Shirley Ho and Marc Huertas-Company and Kartheik G. Iyer and Maja Jablonska and Francois Lanusse and Henry W. Leung and Kaisey Mandel and Juan Rafael Martínez-Galarza and Peter Melchior and Lucas Meyer and Liam H. Parker and Helen Qu and Jeff Shen and Michael J. Smith and Connor Stone and Mike Walmsley and John F. Wu},
  journal= {arXiv preprint arXiv:2412.02527},
  year   = {2024}
}

备注

Accepted at NeurIPS Datasets and Benchmarks track