Supernovae are rare, bright explosions that mark the death of massive stars. Photo from the University of Oxford website
A new AI-based tool has reduced the workload on astronomers by 85% — sifting through thousands of warning data to isolate only those caused by supernovae (powerful explosions of dead stars).
As Delo.ua writes, the results were published in The Astrophysical Journal, which are also described on the website of the University of Oxford.
The study's lead author, Dr Eloise Stevens (Department of Physics, University of Oxford), said: “ The most amazing thing is how little data is needed. Just 15,000 examples and the power of my laptop were enough to train 'smart' algorithms to do the hard work and automate what used to take hours of human labor every day. This proves that, under expert guidance, AI can transform astronomical discoveries without the need for huge data sets or supercomputers . “
Searching for “ needle in a cosmic haystack “
Supernovae are rare, bright explosions that mark the death of massive stars, events that help scientists understand the origins of chemical elements. These explosions appear suddenly in the night sky, and you have to catch them before they fade away — essentially a cosmic game of spot the difference .
A team of researchers led by the University of Oxford and Queen’s University Belfast are looking for such events using the Asteroid Terrestrial Impact Last Alert System (ATLAS). Originally designed as an early warning system for asteroid impacts, ATLAS scans the entire visible sky every 24–48 hours using five telescopes around the world. It is a NASA project led by the University of Hawaii, and Oxford is responsible for processing data on powerful explosions outside our galaxy.
Every night, the system generates millions of potential signals, most of which turn out to be noise (instrumental errors or known objects).
Even after applying standard filtering methods and automated image analysis, researchers were left with 200 to 400 candidates per day that had to be manually screened. Only a few of these turned out to be truly interesting phenomena—supernovae or extragalactic transients (optical analogues of gamma-ray bursts).
“ This manual verification took several hours each day,” added Dr. Stevens. “With this new tool, we can free up scientists' time for what they do best – creative problem-solving and finding answers about the nature of the universe. It's the astrophysical equivalent of having a bot do the laundry while you focus on your art! “
Virtual Research Assistant
The new tool is called the Virtual Research Assistant (VRA) — a set of automated bots that mimic a human decision-making process, ranking signals by the likelihood that they are real extragalactic explosions.
Unlike many AI approaches that require huge training datasets and supercomputers, VRA uses a more “ lightweight “ methodology. Instead of “ voracious “ neural networks, the system uses smaller algorithms based on decision trees that look for patterns in selected data features. This allows scientists to directly input their expertise into the model and guide the algorithms to key features for retrieval.
