Astronomaly at scale: searching for anomalies amongst 4 million galaxies - Contains very interesting objects, lenses, mergers, etc - the ASTRONOMALY Project

Modern astronomical surveys are producing datasets of unprecedented size and richness, increasing the potential for high-impact scientific discovery. This possibility, coupled with the challenge of exploring a large number of sources, has led to the development of novel machine-learning-based anomaly detection approaches, such as Astronomaly. For the first time, we test the scalability of Astronomaly by applying it to almost 4 million images of galaxies from the Dark Energy Camera Legacy Survey. We use a trained deep learning algorithm to learn useful representations of the images and pass these to the anomaly detection algorithm isolation forest, coupled with Astronomaly’s active learning method, to discover interesting sources. We find that data selection criteria have a significant impact on the trade-off between finding rare sources such as strong lenses and introducing artefacts into the dataset. We demonstrate that active learning is required to identify the most interesting sources and reduce artefacts, while anomaly detection methods alone are insufficient. Using Astronomaly, we find 1635 anomalies among the top 2000 sources in the dataset after applying active learning, including 8 strong gravitational lens candidates, 1609 galaxy merger candidates, and 18 previously unidentified sources exhibiting highly unusual morphology. Our results show that by leveraging the human-machine interface, Astronomaly is able to rapidly identify sources of scientific interest even in large datasets.

Paper Access below:

Pictures from the article

Table 2.

Information about the 18 initially unidentified anomalous sources detected in the main set. The second and third columns show the right ascension and declination in degrees, respectively. The fourth column lists any known source names of the target from other surveys or catalogues if they were matched. Blank entries are unmatched sources.

Entry RA
[deg] Declination
[deg] Identifiers
U1 209.7286 29.5764 2MASS J13585490+2934356
U2 342.9047 17.8460 2MASX J22513724+1750457
U3 60.7336 βˆ’15.2434 LEDA 913772
U4 42.2626 3.2043
U5 27.0134 βˆ’21.6656 ESO 543-16
U6 22.1821 βˆ’2.3705
U7 69.8371 βˆ’50.5307 ESO 202-45
U8 31.6507 βˆ’28.0090 2dFGRS TGS226Z042
U9 50.7173 βˆ’11.8873
U10 310.1343 1.8177
U11 63.0540 βˆ’24.9667
U12 36.2380 βˆ’27.2902 2dFGRS TGS230Z092
U13 210.4254 33.8215 NVSS J140141+334937
U14 212.8158 0.1169 SDSS J141116.31+000654.9
U15 46.9290 βˆ’14.1055 LEDA 928927
U16 55.5332 βˆ’19.8353
U17 106.5698 68.5793
U18 32.2221 βˆ’0.9785 SDSS J020853.45-005841.1

Table 3.

The gravitational lens candidates that have been identified in the top 10 000 anomalies. The second and third columns show the right ascension and declination in degrees, respectively. The last column indicates whether the candidates have been confirmed to be a lens, a lens candidate or a candidate that has not been matched to any other catalogue yet.

Entry RA
[deg] Declination
[deg] Information
L1 35.2352 βˆ’7.7199 Confirmed lens – [More et al. ([2012](javascript:;))]
L2 27.9503 βˆ’32.6199 Candidate – [Jacobs et al. ([2019](javascript:;))]
L3 61.6016 βˆ’26.7733 Candidate – [Jacobs et al. ([2019](javascript:;))]
L4 128.1546 13.5797 Candidate – [Shu et al. ([2017](javascript:;))]
L5 340.2492 βˆ’52.7542 Candidate – [Diehl et al. ([2017](javascript:;))]
L6 78.3564 βˆ’30.8416 Candidate – Previously undetected
L7 9.7307 7.3230 Candidate – Previously undetected
L8 60.1041 βˆ’16.3973 Candidate – Previously undetected

Entry RA
[deg] Declination
[deg] Entry RA
[deg] Declination
M1 11.4865 32.2940 M25 333.0850 0.5605
M2 12.0642 βˆ’25.6886 M26 333.2532 21.9719
M3 12.4649 17.7756 M27 335.2230 13.4425
M4 123.6086 37.2613 M28 36.9431 26.5896
M5 127.6318 18.2050 M29 40.4523 βˆ’49.0052
M6 134.2013 30.8611 M30 42.9094 βˆ’16.6571
M7 138.9657 13.5864 M31 43.6074 βˆ’64.1671
M8 147.5532 βˆ’5.6929 M32 44.8704 βˆ’14.2910
M9 168.9463 15.8239 M33 46.3564 βˆ’19.4730
M10 192.1130 15.5824 M34 49.0504 βˆ’12.1633
M11 205.4201 13.5041 M35 50.3167 βˆ’28.3164
M12 213.6448 24.4104 M36 60.4281 βˆ’23.5613
M13 22.8853 βˆ’22.3706 M37 63.3138 βˆ’14.6525
M14 222.3588 23.3494 M38 64.5453 βˆ’31.2488
M15 229.9712 2.6200 M39 68.7329 βˆ’40.0342
M16 239.7994 20.7477 M40 70.2079 βˆ’44.9443
M17 24.0000 βˆ’11.7047 M41 72.2531 βˆ’33.3201
M18 25.4929 βˆ’13.8470 M42 76.0434 βˆ’57.2689
M19 251.9828 10.5277 M43 79.3557 βˆ’48.3955
M20 254.6314 58.9370 M44 80.1525 βˆ’38.6625
M21 28.6783 27.3285 M45 81.0061 βˆ’32.2876
M22 30.6765 βˆ’2.5885 M46 81.8048 βˆ’31.4080
M23 318.6995 βˆ’2.1876 M47 83.3110 βˆ’39.4453
M24 330.9685 βˆ’1.6943 M48 85.2283 βˆ’20.3304

1 Like

Wow! Congrats to these guys for an interesting sample! Surely I can’t code lol and it will be based upon human

1 Like