
Machine-learning sweep of Hubble data reveals 800+ new cosmic curiosities
AnomalyMatch scanned roughly 100 million Hubble image cutouts spanning ~35 years, returning 1,255 unique objects across 18 classes, with more than 800 not previously described in the literature. Most are known categories (merging galaxies, lenses, rings, jellyfish galaxies), but a few dozen resist current schemes. The study shows AI can scale anomaly discovery for future large surveys (Euclid, Rubin Observatory), while confirming that human inspection remains essential and follow-up observations are needed to validate lens candidates and classify unconfirmed objects.












