A team of astronomers has used a new artificial intelligence assisted technique to uncover rare astronomical phenomena in archived data from NASA's Hubble Space Telescope. They sifted through nearly 100 million small image cutouts from the Hubble Legacy Archive, each only a few dozen pixels across, and identified more than 1,300 unusually shaped objects in just two and a half days, including ove Los Angeles CA (SPX) Jan 29, 2026
A team of astronomers has used a new artificial intelligence assisted technique to uncover rare astronomical phenomena in archived data from NASA's Hubble Space Telescope. They sifted through nearly 100 million small image cutouts from the Hubble Legacy Archive, each only a few dozen pixels across, and identified more than 1,300 unusually shaped objects in just two and a half days, including over 800 that had never been documented in scientific literature.
Most of the newly cataloged anomalies are galaxies in the midst of mergers or interactions that display distorted shapes or long, trailing streams of stars and gas. Others are examples of gravitational lensing, where the gravity of a foreground galaxy warps spacetime and bends light from a background galaxy into bright arcs or nearly complete rings. The search also revealed galaxies packed with massive star forming clumps, jellyfish like systems with gaseous tentacles, and edge on planet forming disks in the Milky Way that resemble hamburgers, along with several dozen objects that did not fit any existing classification.
The explosion of data from Hubble and other observatories has made it increasingly difficult to find rare or unexpected objects with traditional methods. Hubble's three and a half decades of observations have produced a vast archive, and never before have astronomers had access to such a large volume of sky images that might hide subtle or unusual phenomena. Manually inspecting every image in this archive is no longer practical, even for large teams of experts.
To tackle this challenge, David O'Ryan and Pablo Gomez of ESA developed an AI tool designed to scan millions of astronomical images far faster than human experts can. Their neural network, called AnomalyMatch, was trained to recognize patterns and identify rare or unusual sources, in a way that mimics how the human brain processes visual information but at a much greater scale. By automating the first pass over the data, the system filters out the most promising anomalies for astronomers to examine in detail.
When AnomalyMatch was applied to the full Hubble Legacy Archive, it marked the first systematic search for astrophysical anomalies across the entire dataset. The algorithm produced a ranked list of candidates, and the researchers then manually inspected the highest rated sources to verify their nature. This combined AI and human review process confirmed more than 1,300 true anomalies, demonstrating that the archive still holds many surprises despite decades of intensive study.
Citizen science projects and visual inspections by professional astronomers have previously identified some unusual Hubble images, but they cannot keep pace with the growing volume of data. The new results show that AI can complement these efforts by efficiently surfacing rare phenomena that might otherwise remain buried in the archive. As Gomez noted, the discovery of so many previously undocumented anomalies in Hubble data highlights how tools like AnomalyMatch can significantly enhance the scientific return from existing observations.
Hubble is only one of several observatories whose archives are expected to benefit from AI driven analysis. NASA's upcoming Nancy Grace Roman Space Telescope, ESA's Euclid mission, and the National Science Foundation and Department of Energy's Vera C. Rubin Observatory will all produce enormous datasets that far exceed Hubble's in size. Automated anomaly detection will be essential for navigating this data deluge, helping astronomers quickly pinpoint unusual objects and phenomena that merit follow up.
The success of AnomalyMatch hints at the kinds of discoveries that may emerge as these next generation facilities begin operations. By rapidly scanning images for outliers while astronomers focus on interpretation, AI tools can open new windows on the universe and possibly reveal types of objects that have never been seen before. The approach also shows how archived observations remain a rich resource for discovery when paired with modern data analysis techniques.
The Hubble Space Telescope itself has been operating for more than three decades and continues to reshape our understanding of the cosmos. It is a project of international cooperation between NASA and ESA, with NASA's Goddard Space Flight Center in Greenbelt, Maryland managing the observatory and mission operations. Lockheed Martin Space in Denver supports operations at Goddard, and the Space Telescope Science Institute in Baltimore, operated by the Association of Universities for Research in Astronomy, conducts Hubble science operations for NASA.
Related Links
Hubble Telescope
Stellar Chemistry, The Universe And All Within It