Two researchers at the European Space Agency have used an artificial intelligence tool to comb through nearly 100 million image cutouts from NASA’s Hubble Space Telescope archive in about two and a half days, identifying more than 1,300 unusual objects, more than 800 of which had never been described in the scientific literature, NASA and ESA announced on January 27.
The tool, a neural network called AnomalyMatch, was developed by ESA scientists David O’Ryan and Pablo Gómez. It was trained to recognize rare objects such as gravitational arcs and so-called jellyfish galaxies by learning visual patterns in the data, and then applied to the full Hubble Legacy Archive, which spans 35 years of observations. According to NASA, it is the first systematic search for anomalies across the entire dataset.
What the search turned up
Most of the flagged objects were galaxies in the middle of mergers or interactions, showing distorted shapes or long trailing streams of stars and gas. Others were gravitational lenses, in which the gravity of a foreground galaxy bends the light of a more distant one into arcs or rings. The results also included galaxies with massive star-forming clumps, jellyfish galaxies trailing gaseous tentacles, and edge-on planet-forming disks within our own galaxy. Several dozen objects did not fit any existing classification scheme.
The ESA release breaks the haul down further, listing 138 candidate gravitational lenses, 417 galaxy mergers and 18 jellyfish galaxies among the confirmed anomalies. After the algorithm ranked likely candidates, the researchers reviewed the top-rated sources by hand and confirmed more than 1,300 as genuine. Each cutout measured only a few dozen pixels, or 7 to 8 arcseconds, on a side.
“Archival observations from the Hubble Space Telescope now span 35 years, offering a rich dataset in which astrophysical anomalies may be hidden,” said O’Ryan, lead author of the study, which was published in the journal Astronomy and Astrophysics.
Built for the coming data flood
Unusual objects in astronomy have traditionally been found by chance or by astronomers inspecting images one at a time, with citizen science projects extending that reach. NASA notes that even those efforts cannot keep pace with archives as large as Hubble’s, let alone the output of wide-field survey missions.
That is where the researchers see the tool’s main value. ESA’s Euclid mission, NASA’s Nancy Grace Roman Space Telescope and the Vera C. Rubin Observatory, funded by the US National Science Foundation and Department of Energy, are expected to produce unprecedented volumes of imagery over the coming years. “This is a powerful demonstration of how AI can enhance the scientific return of archival datasets,” Gómez said, adding that the number of undocumented anomalies found in Hubble data points to the tool’s potential for future surveys.
