More time for research: AI helps in the search for stellar explosions
Thanks to an AI system, astronomers no longer have to spend a lot of time looking for possible finds. The tool imitates human decision-making.
(Image: muratart/Shutterstock.com)
A research group in the UK has developed an AI tool that can automatically search astronomical images for signals of a supernova, saving experts several hours of work every day. The University of Oxford has now announced this and explained that the tool has already proven its efficiency over the course of a year. It consists of a series of algorithms that mimic human decision-making and does not rely on large amounts of training data or supercomputers. Within a year, the Virtual Research Assistant (VRA) has ensured that 85 percent fewer images need to be checked by a human and less than 0.08 percent of real signals are overlooked.
Like a domestic robot
As the research group explains, the AI tool automatically searches images from the Asteroid Terrestrial Impact Last Alert System (ATLAS), which was set up to search for potentially dangerous asteroids and has discovered the interstellar comet 3I/ATLAS, among others. Every night it delivers millions of potential finds, but the vast majority of them are noise. Even after applying various filters and corrections, there are still around 200 to 400 candidates for a signal that could come from a stellar explosion. These then have to be checked manually, which takes several hours every day. In the end, about a handful of “really interesting phenomena” remain, including optical signals for gamma-ray bursts in addition to supernovae.
Videos by heise
The AI tool “is the astrophysical equivalent of a robot that does your laundry so you can create art,” says Héloïse Stevance from the University of Oxford, who led the development. By relieving astronomers of tedious control work, it gives them time for creative problem-solving and research itself. Since the end of 2024, the VRA has even been directly connected to a telescope in South Africa and, if necessary, can initiate follow-up observations of a particularly exciting signal before a person has even been involved. Supernovae have already been confirmed in this way. The team presents its work in the Astrophysical Journal.
(mho)
