Universe Today
AI faces daily criticism from people worried about its ill-effects. But the type of AI that draws this ire are Large Language Models (LLMs). There are other types of AI with specialized functions that don't make it onto the front pages. Combing through vast troves of astronomical data is a perfect task for AI that is unlikely to be replicated by human minds.
A case in point is in new research published in Astronomy and Astrophysics. It's titled "Identifying astrophysical anomalies in 99.6 million source cutouts from the Hubble legacy archive using AnomalyMatch," and the authors are David O'Ryan and Pablo Gomez. O'Ryan and Gomez are both from the ESA's European Space Astronomy Centre (ESAC) in Madrid, Spain.
Our powerful collection of astronomical telescopes are creating a mass of data. The JWST contributes about 57 GB of data every day, depending on what observations are scheduled. The Vera Rubin Observatory, with the largest digital camera ever built, will vastly outpace that. It will generate about 20 terabytes of raw data each night and requires special infrastructure just to handle it. With powerful new telescopes like the Giant Magellan Telescope and Extremely Large Telescope coming online soon, the amount of astronomical data needing scientific scrutiny is growing into a deluge.
These vast quantities of data are bound to hold many hidden surprises. Our technology has outpaced the capacity of organic brains to process it all. But technological AI is catching up to astronomy's mass data-generation capability.
“Archival observations from the Hubble Space Telescope now stretch back 35 years, providing a treasure trove of data in which astrophysical anomalies might be found,” said co-lead author O’Ryan.
"Astronomical archives contain vast quantities of unexplored data that potentially harbour rare and scientifically valuable cosmic phenomena," the authors write. "We leverage new semi-supervised methods to extract such objects from the Hubble Legacy Archive."
Astrophyscial anomalies are important because they can be outliers that present a different side of nature. A trained scientist might be attuned to them and find them relatively easy. But there's just too much data.
The researchers used a recently-developed anomaly detection framework named AnomalyMatch to rapidly search through almost 100 million image cutouts from the Hubble Legacy Archive. The archive contains images going back aboutt 35 years.
AnomalyMatch is different AI than the type the techno-oligarchs are trying to cram into every piece of consumer software. It's a neural network, a machine learning tool inspired by the human brain. "AnomalyMatch is tailored for large-scale applications, efficiently processing predictions for ≈100 million images within three days on a single GPU," the authors wrote in a previous paper that presented the AnomalyMatch tool.
It took AnomalyMatch only 2 to 3 days to process this much data, a fraction of the time it would take human minds. It's the first time the Hubble Legacy Archive has undergone such a systematic search for anomalies. AnomalyMatch generated a list of likely anomalies. That list contained almost 1,400 anomalous objects, a number that's handled much more easily by human minds. O'Ryan and Gomez went through these 1,400 objects manually and determined that 1,300 of them were in fact anomalies, and that more than 800 of them have never been documented.
Merging and interacting galaxies were the most common type of anomaly detected in the Archive. There were 417 of them.
This group of gravitationally interacting galaxies is one of the anomalous the researchers found in the Hubble Legacy Archive. The distorted shapes and tidal tails illustrate the gravitational effects. Image Credit: ESA/Hubble & NASA, D. O’Ryan, P. Gómez (European Space Agency), M. Zamani (ESA/Hubble)
The researchers also found 86 new potential gravitational lenses. These are important because they bring objects that are otherwise too distant to observe into reach. They also help scientists study the distribution of dark matter in the Universe, measure distances and cosmic expansion, and test general relativity. "We identify many gravitational lenses that are already identified in the literature – but many candidate new lenses," the authors write.
This is one of the gravitational lenses found in the Hubble Legacy Archive. The reddish elliptical galaxy is the foreground lens and a blue spiral galaxy in the background is magnified and distorted by the elliptical galaxy. These types of alignments bring distant objects into observational reach. Image Credit: ESA/Hubble & NASA, D. O’Ryan, P. Gómez (European Space Agency), M. Zamani (ESA/Hubble)
There were other anomalies in the Archive, too. AnomalyMatch found other rare objects like jellyfish galaxies. These are found in galaxy clusters where ram pressure is stripping gas from the galaxy, leaving a long tail lit up with star formation. There were 35 of them found in the Archive.
The research also turned up some anomalies with uncertain natures. One of them is a strange sight, a galaxy with a swirling core and open lobes.
This galaxy highlights the anomalous nature of some difficult-to-categorize objects. It's a bi-polar galaxy with a compact swirling core and an open lobe at each side. This object was newly-discovered and previously unknown. It's not clear what type of galaxy it is, and if it's strange morphology is related to a merger. Its discovery highlights the utility of AI tools to search through astronomical archives. Image Credit: ESA/Hubble & NASA, D. O’Ryan, P. Gómez (European Space Agency), M. Zamani (ESA/Hubble)
Finding hidden surprises in vast quantities of astronomical data is an admirable use of AI. Along with the previously mentioned anomalies, the researchers also uncovered overlapping galaxies, clumpy galaxies, ring galaxies, and even high-redshift galaxies so close to detection limits they're difficult to discern. They also found jetted galaxies and AGN-hosting galaxies.
*This figure from the research shows five examples of every anomaly sub-class for which we found at least five objects, not including lensed quasars. These were selected as representative of each sub-class. Image Credit: O'Ryan and Gomez 2026. A&A*
If all astronomical observations stopped tomorrow, the discoveries wouldn't stop. Capable AI tools are destined to become more and more powerful. Massive existing datasets from the Hubble and from other missions like the ESA's Gaia are feeding grounds for future tools.
Who knows what's waiting to be discovered in all that data?
“This is a powerful demonstration of how AI can enhance the scientific return of archival datasets,” Gómez said. “The discovery of so many previously undocumented anomalies in Hubble data underscores the tool’s potential for future surveys.”
