Practical galaxy morphology tools from deep supervised representation learning

Journal Article · Mon Feb 28 00:00:00 EST 2022 · Monthly Notices of the Royal Astronomical Society

DOI:https://doi.org/10.1093/mnras/stac525· OSTI ID:1982680

Walmsley, Mike ^[1]; Scaife, Anna M. ^[2]; Lintott, Chris ^[3]; Lochner, Michelle ^[4]; Etsebeth, Verlon ^[5]; Géron, Tobias ^[3]; Dickinson, Hugh ^[6]; Fortson, Lucy ^[7]; Kruk, Sandor ^[8]; Masters, Karen L. ^[9]; Mantha, Kameswara Bharadwaj ^[7]; Simmons, Brooke D. ^[10]

---

1. University of Manchester (United Kingdom)
2. University of Manchester (United Kingdom); Alan Turing Institute, London (United Kingdom)
3. University of Oxford (United Kingdom)
4. University of the Western Cape, Cape Town (South Africa); South African Radio Astronomy Observatory (SARAO), Cape Town (South Africa)
5. University of the Western Cape, Cape Town (South Africa)
6. The Open University, Kents Hill (United Kingdom)
7. University of Minnesota, Minneapolis, MN (United States)
8. Max-Planck-Institut fur extraterrestrische Physik, Garching bei München, (Germany); European Space Agency, Noordwijk (Netherlands)
9. Haverford College, PA (United States)
10. Lancaster University, Bailrigg (United Kingdom)

Astronomers have typically set out to solve supervised machine learning problems by creating their own representations from scratch. We show that deep learning models trained to answer every Galaxy Zoo DECaLS question learn meaningful semantic representations of galaxies that are useful for new tasks on which the models were never trained. We exploit these representations to outperform several recent approaches at practical tasks crucial for investigating large galaxy samples. The first task is identifying galaxies of similar morphology to a query galaxy. Given a single galaxy assigned a free text tag by humans (e.g. ‘#diffuse’), we can find galaxies matching that tag for most tags. The second task is identifying the most interesting anomalies to a particular researcher. Our approach is 100 per cent accurate at identifying the most interesting 100 anomalies (as judged by Galaxy Zoo 2 volunteers). The third task is adapting a model to solve a new task using only a small number of newly labelled galaxies. Models fine-tuned from our representation are better able to identify ring galaxies than models fine-tuned from terrestrial images (ImageNet) or trained from scratch. We solve each task with very few new labels; either one (for the similarity search) or several hundred (for anomaly detection or fine-tuning). This challenges the longstanding view that deep supervised methods require new large labelled data sets for practical use in astronomy. To help the community benefit from our pretrained models, we release our fine-tuning code zoobot. Zoobot is accessible to researchers with no prior experience in deep learning.

• Research Organization: Lawrence Berkeley National Laboratory (LBNL), Berkeley, CA (United States). National Energy Research Scientific Computing Center (NERSC); University of Minnesota, Minneapolis, MN (United States)
• Sponsoring Organization: Alan Turing Institute; Alfred P. Sloan Foundation; National Research Foundation (NRF); National Science Foundation (NSF); South African Radio Astronomy Observatory; USDOE Office of Science (SC), High Energy Physics (HEP)
• Grant/Contract Number: AC02-05CH11231
• OSTI ID: 1982680
• Journal Information: Monthly Notices of the Royal Astronomical Society, Journal Name: Monthly Notices of the Royal Astronomical Society Journal Issue: 2 Vol. 513; ISSN 0035-8711
• Publisher: Oxford University Press
• Country of Publication: United States
• Language: English

Related Subjects

79 ASTRONOMY AND ASTROPHYSICS
data analysis
evolution
galaxies
general
methods
public release
software