AI uncovers 800+ hidden cosmic anomalies in Hubble archive
ESA researchers trained a neural network called AnomalyMatch to sift through nearly 100 million image cutouts from the Hubble Legacy Archive in about 2.5 days, flagging 1,400 potential anomalies. After human review, more than 800 of these were undocumented, including merging galaxies, gravitational lenses, planet-forming disks seen edge-on, and other unusual objects, showcasing AI’s power to process vast astronomical datasets while still requiring expert confirmation.