AI scans 100M Hubble images in 2.5 days, finds 1,300 cosmic oddities

An artificial intelligence system has combed through nearly 100 million archival images from the Hubble Space Telescope in just two and a half days, identifying 1,300 unusual cosmic objects. Of those, more than 800 had never been described in any scientific paper before. The findings were reported by researchers who developed the AI to scan decades of Hubble data for anomalies that human eyes might have missed.

The survey covered images taken between 2002 and 2022, a period spanning 20 years of observations. The AI flagged objects that stood out from typical galaxies, stars, or nebulae, including potential gravitational lenses, interacting galaxies, and other rare phenomena. Astronomers say the technique could dramatically speed up the discovery of new celestial features in vast astronomical archives.

Why this matters is that Hubble has produced millions of images over its 30-plus-year mission, but only a fraction have been thoroughly examined for unusual objects. The AI's ability to process such a large dataset in under three days shows how machine learning can help scientists find needles in a cosmic haystack. The 1,300 candidates include some that may represent entirely new classes of objects or rare evolutionary stages.

The researchers now plan to follow up on the most promising candidates using ground-based telescopes and possibly the James Webb Space Telescope. They also aim to apply the same AI technique to other large surveys, such as those from the upcoming Nancy Grace Roman Space Telescope. The work demonstrates a new way to mine existing data for discoveries that were previously overlooked.