There are 800 previously unpublished ‘astrophysical anomalies’ in the Hubble archives

A new tool artificial intelligence developed atEuropean Space Agency (ESA) allowed us to uncover over 800 “astrophysical anomalies” up to now not documented in the archives of Hubble Space Telescopesystematically sifting through a huge mass of images in just a few days. The work is signed by researchers David O’Ryan and Pablo Gómez and was published in the journal Astronomy & Astrophysics.

The team trained a neural network call AnomalyMatch using theHubble Legacy Archivean archive that collects observations accumulated over 35 years. The goal is to identify rare items and “out of the norm”, as colliding or merging galaxies, gravitational lenses, ring galaxies and other unusual structures, which to the human eye are difficult to find such vast datasets.

According to theESA, AnomalyMatch he analyzed almost 100 million “snips” of images of the archive, completing the search in approximately two and a half days. The results with the higher probability of being anomalous they were then checked manually by researchers: the check led to further identification 1,300 real anomaliesof which more than 800 were not present in the scientific literature.

Most cases concern interacting galaxies con distorted shapes e code di stelle e gas. There are many others gravitational lensesin which the severity of a galaxy in the foreground bends it space-time and warps light from a more distant galaxy into arcs or rings. Among the anomalies also appear “jellyfish galaxies” with “tentacles” of gas, galaxies with large clusters of stars and protoplanetary disks cut visas; in some cases objects have emerged that do not fit into known categories.

For theESA the result shows the potential of AI in “maximizing” theoutput scientifico Of large archives and in dealing with the new wave of data arriving from missions and observatories such as Euclidil Vera C. Rubin Observatory and the future Nancy Grace Roman Space Telescope.