AI Discovers 1,300 Anomalies in Hubble's Legacy Archive Using Pattern Recognition
An AI model has combed through 35 years of Hubble Space Telescope data to uncover 1,300 previously undetected cosmic anomalies. The European Space Agency (ESA) developed AnomalyMatch, a pattern recognition system designed to analyze Hubble's 1.7 million archived images.
By processing 100 million image cutouts in three days, the system identified 100+ undocumented objects alongside known phenomena like merging galaxies and star-forming clumps.
"This is a powerful demonstration of how AI can enhance the scientific return of archival datasets," stated Pablo Gómez of ESA. The study, published in Astronomy and Astrophysics in December 2025, highlights the system's ability to detect rare astrophysical events.
Among the anomalies were 'jellyfish galaxies'—galaxies with trailing stellar streams—and 'hamburger' disks, flattened structures with central bulges. The classification of these objects remains an active area of research, as their formation mechanisms are not yet fully understood.
David O'Ryan, lead author of the study, noted that Hubble's 35-year observational record provides a unique temporal baseline. The team emphasized that while AI accelerates discovery, human validation remains critical for interpreting complex astrophysical phenomena.
The methodology involved training AnomalyMatch on known galaxy types before applying it to the full archive, with results manually reviewed to confirm authenticity.
Researchers caution that the system's effectiveness depends on the quality of training data. "Archival observations from the Hubble Space Telescope now stretch back 35 years," O'Ryan explained, "but challenges remain in distinguishing transient events from instrumental artifacts." The study authors advocate for continued refinement of machine learning approaches to improve detection accuracy in future missions.
