Trust most of the work to artificial intelligence to save time. This is what researchers at the University of Oxford (United Kingdom) and Queen’s in Belfast (Northern Ireland) have done to detect supernovae. Marking the death of mass stars, these rare and bright explosions help scientists understand the origin of chemical elements.
The problem is that they seem unexpectedly in heaven and that they must identify quickly before fading. However, several hours are necessary to verify the signals identified by the system developed by the researchers, who are potentially linked to supernovae.
Called Atlas (for the last terrrescial impact alert system of the asteroids, that is, a warning system for terrestrial asteroids in French), the latter analyzes the entire visible sky every 24 to 48 hours using five telescopes distributed throughout the planet. The Oxford University then processes data related to high intensity explosions beyond the Milky Way. The investigation that leads to millions of potential alerts every night, most of which are due to noise. Another said that these are instrumental errors or known objects.
A reduced workload of approximately 85%
From there, scientists use automated filtering techniques and image analysis to reduce the number of alerts, but this is not yet enough. Because they end with 200 to 400 signals to classify manually, of which only a handful are “really interesting phenomena”, such as supernovae.
But the AI tool developed by scientists allowed them to further reduce the workload. Called Virtual Research Assistant (TRU), it is a set of automated robots that mimic the human decision -making process classifying alerts according to their probability of being real extragalactic explosions.
Using algorithms for this, it managed to filter more than 30,000 alerts during its first year of use, without less than 0.08% of the real supernova signals. The number of alerts transmitted to human observers for verification was reduced by around 85%, with more than 99.9% of them that would probably be supernova.
The Real also updates its evaluation every time a telescope reviews the same sky zone. Therefore, each signal is automatically revived and reassess for several nights, and are only the most promising that are transmitted to human astronomers.
Detect everything
Since December 2024, the AI tool has been connected to the Lesedi South African telescope to automatically trigger follow -up observations for the most promising signals, which has confirmed new supernovae. And researchers do not plan to stop there.
At the beginning of 2026 the Legacy of Space and Time (LSST) survey of the Verra Rubin Observatory (Chile) will be launched. A system that will study the entire sky of the southern hemisphere at intervals of a few days, generating more than 500 image and data petacts.
At the same time, it also develops tra for British and European LSST data corridors, hoping to use this information to create robots capable of preventing preventively supernovae when predicting the moment and the place when they explode.
Source: BFM TV
