The universe is filled with innumerable astrophysical objects, each one different from the last. But even amid this vast diversity, some stand out as truly bizarre.
A pair of astronomers recently discovered hundreds of these cosmic weirdos buried in archival Hubble Space Telescope data. These objects have waited years for researchers to catalog and investigate their unusual characteristics, and thanks to AI, they finally have.
“Archival observations from the Hubble Space Telescope now span 35 years, offering a rich dataset in which astrophysical anomalies may be hidden,” David O’Ryan, a research fellow at the European Space Agency (ESA) and lead author of the study published in Astronomy & Astrophysics, said in an agency statement.
O’Ryan and his colleague, ESA data scientist Pablo Gómez, created an AI-assisted data analysis tool called AnomalyMatch and used it to search for rare astronomical objects in the Hubble Legacy Archive. It took just two and a half days to sift through nearly 100 million image cutouts and identify nearly 1,400 anomalous objects, 800 of which were previously unknown to science.
“This is a powerful demonstration of how AI can enhance the scientific return of archival datasets,” Gómez said in a NASA statement. “The discovery of so many previously undocumented anomalies in Hubble data underscores the tool’s potential for future surveys.”
Mining Hubble’s vast archive
Hubble has spent more than three decades continuously surveying the cosmos. To date, the telescope has made more than 1.7 million observations, building a data goldmine that has significantly expanded our understanding of the universe.
However, sifting through this mountain of data to find rare and anomalous objects, such as colliding galaxies, gravitational lenses, and ring galaxies, is an onerous task for astronomers. Gómez and O’Ryan developed AnomalyMatch to do the heavy lifting for them.
Their AI tool is a neural network—a machine learning model designed to mimic the way the human brain processes data and recognizes patterns. AnomalyMatch is trained to sniff out cosmic objects that look unusual, compiling a list of targets that astronomers like O’Ryan and Gómez can then examine more closely to confirm and classify.
A wealth of weirdos
Of the 800-odd oddballs AnomalyMatch and its creators identified, most were galaxies actively merging or interacting with other galaxies, morphing them into unusual shapes or giving them trailing tails of stars and gas.
They also found many gravitational lenses—massive celestial bodies that bend spacetime and warp the light around them, acting as a natural lens—and other rare objects such as galaxies with huge star clumps, jellyfish galaxies with gaseous “tentacles,” and planet-forming disks that resemble hamburgers or butterflies when viewed edge-on.
Most intriguing were several dozen objects that defied classification entirely, presenting new opportunities to probe never-before-seen cosmic structures.
The findings show that neural networks like AnomalyMatch can maximize the value of data archives like Hubble’s. Gómez and O’Ryan hope their tool will unlock new discoveries from forthcoming datasets as well, including that of ESA’s Euclid space telescope and the National Science Foundation and U.S. Department of Energy’s Vera C. Rubin Observatory.
These next-generation surveys will produce a deluge of data, and analyzing that data will require next-generation techniques. Combing through the cosmos with AI could open the door to a whole new world of scientific discovery.
