Self-supervised Representation Learning for Astronomical Images

Journal Article · Mon Apr 26 00:00:00 EDT 2021 · The Astrophysical Journal. Letters

DOI:https://doi.org/10.3847/2041-8213/abf2c7· OSTI ID:1813371

Hayat, Md Abul ^[1]; Stein, George ^[2]; Harrington, Peter ^[3]; Lukić, Zarija ^[3]; Mustafa, Mustafa ^[3]

---

1. Univ. of Arkansas, Fayetteville, AR (United States); Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States)
2. Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States); Univ. of California, Berkeley, CA (United States)
3. Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States)

Sky surveys are the largest data generators in astronomy, making automated tools for extracting meaningful scientific information an absolute necessity. We show that, without the need for labels, self-supervised learning recovers representations of sky survey images that are semantically useful for a variety of scientific tasks. These representations can be directly used as features, or fine-tuned, to outperform supervised methods trained only on labeled data. We apply a contrastive learning framework on multiband galaxy photometry from the Sloan Digital Sky Survey (SDSS), to learn image representations. We then use them for galaxy morphology classification and fine-tune them for photometric redshift estimation, using labels from the Galaxy Zoo 2 data set and SDSS spectroscopy. In both downstream tasks, using the same learned representations, we outperform the supervised state-of-the-art results, and we show that our approach can achieve the accuracy of supervised models while using 2-4 times fewer labels for training. The codes, trained models, and data can be found at https://portal.nersc.gov/project/dasrepo/self-supervised-learning-sdss.

• Research Organization: Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States)
• Sponsoring Organization: USDOE Office of Science (SC)
• Grant/Contract Number: AC02-05CH11231
• OSTI ID: 1813371
• Alternate ID(s): OSTI ID: 23154197
• Journal Information: The Astrophysical Journal. Letters, Journal Name: The Astrophysical Journal. Letters Journal Issue: 2 Vol. 911; ISSN 2041-8205
• Publisher: IOP Publishing
• Country of Publication: United States
• Language: English

Related Subjects

79 ASTRONOMY AND ASTROPHYSICS
astronomical methods
computational methods
observational astronomy
observational cosmology
sky surveys