Astronomy: AI automatically detects, confirms and classifies supernovae

For the first time, a new artificial intelligence system has detected, confirmed, classified a supernova and shared the data with the astronomy community without human intervention. This is what was stated in a report issued by Northwestern University in the United States of America, where the system was developed. Thanks to a now-introduced tool called BTSbot (Bright Transient Survey Bot), it will soon no longer be necessary for people to spend time examining and classifying images of suspected starbursts. The system can also independently order the necessary follow-up notes. This would completely remove people from the process, where AI has so far played a supporting role.

advertisement

People can sleep

As the research group led by astronomer Adam Miller explains,Automation in the search for supernovas so far has been used mainly ,initially. Automatic telescopes repeatedly image parts of the night sky and search for new light sources. However, the candidates are then given to people who examine them manually and, if necessary, make follow-up observations using other telescopes to determine the body’s composition by spectrophotometry. The robot is now doing all this, and when it found a supernova called SN 2023tyk (ZTF23abhvlji), it proved that it works. This means that people can now devote themselves to other tasks.

The technology detected the star’s explosion on October 5 in two days old data from the Zwicky Transient Facility (ZTF). A follow-up observation of the Palomar Observatory was automatically performed to determine the spectrum. It was then classified by another AI. Therefore, it is a type Ia supernova. The results were automatically published on October 7. At first, people were nervous, says one of the co-researchers, but when the classification arrived, it brought them great relief. The nice thing is that you can now go to sleep and check in the morning what the AI ​​has identified. The team explains that BTSbot was trained using 1.4 million images of 16,000 astronomical objects.

(meh)