AI finds over 800 anomalies in Hubble photos, dozens defy explanation
A neural network has discovered over 800 anomalies in old Hubble images, primarily unusual galaxies. Dozens of findings defy explanation.
(Image: ESA/Hubble & NASA, D. O’Ryan, P. Gómez (European Space Agency), M. Zamani (ESA/Hubble))
A specially developed AI algorithm has discovered about 1400 anomalies in almost 100 million images from the Hubble Space Telescope. 800 of them were not previously described, and several dozen cannot currently be explained. The European Space Agency ESA has now announced this and published several examples. The unusual astronomical objects are primarily galaxies whose unusual appearance is due to them currently merging with each other, or gravitational lenses. In addition, there are galaxies with gigantic gas clumps, or “tentacles,” of gas, as well as planet-forming dust disks that directly resemble a hamburger when viewed from the side.
Anomalien aus dem Hubblle-Archiv (6 Bilder)
ESA/Hubble & NASA, D. O’Ryan, P. Gómez (European Space Agency), M. Zamani (ESA/Hubble)
)Important tool for expected data volumes
The neural network named “AnomalyMatch” was developed by ESA researchers David O’Ryan and Pablo Gómez. The two explain that such astronomical anomalies are of immense interest for research but are difficult to find in the rapidly growing amount of image data. This is precisely why they trained the algorithm to search for such objects. For the almost 100 million Hubble images, the technique only took two and a half days. Those objects to which “AnomalyMatch” assigned the highest probability of an anomaly were then manually reviewed by the two. They presented the results in the journal Astronomy & Astrophysics.
Videos by heise
This is a “fantastic use” of AI to maximize the scientific value of Hubble's work, Gómez says. It also shows how useful the tool could be for other large datasets. They point out that the most modern instruments, like the Vera C. Rubin Observatory, will soon deliver more than 50 petabytes of scientific data. This data must be systematically searched, and AI can be a great help. The researchers present several of the algorithm's findings, which are primarily unusually looking galaxies. Only one of the unexplainable objects is among them; it is a galaxy with giant structures whose nature is still unclear.
(mho)
