The neural network of two ESA researchers has isolated nearly 1,400 atypical objects in the Hubble archive, more than 800 of which had never been described.
David O'Ryan and Pablo Gómez, from the European Space Agency, built a neural network called AnomalyMatch to identify rare objects in astronomical images. Applied to the Hubble Legacy Archive, the tool processed nearly 100 million vignettes in two and a half days, where a human examination would have taken an incomparable amount of time.
The harvest amounts to nearly 1,400 atypical objects, of which more than 800 had never been documented. The majority are merging or interacting galaxies, with unusual morphologies, sometimes accompanied by long filaments of stars and torn gas. There are also gravitational lenses, where the mass of a galaxy in the foreground bends the light of a background galaxy in arcs or rings, galaxies with large clusters of star formation, "jellyfish" galaxies with gaseous tentacles, and protoplanetary disks seen edge-on.
The network was trained to recognize patterns in the manner of human visual processing. The exercise is the first systematic search for astrophysical anomalies conducted on this entire dataset.
Read the full article on the original source
Go to the source →