The night sky looks like a scattering of identical sparks. But hidden among them are true giants — red supergiants. These stars are hundreds of times wider than the Sun and shine so brightly that they can be seen from enormous distances. You'd think there'd be no better beacons for a map of the galaxy. But until recently, astronomers knew only a few hundred such objects — too few to outline the spiral arms.

The situation changed with the Gaia space telescope. It captures spectra—'rainbow fingerprints' of light — for hundreds of millions of stars. These fingerprints encode temperature, chemical composition, and other properties. Processing such a mountain of data by hand is impossible, so scientists trained a neural network. It scanned each star's spectrum and estimated the probability that it was a red supergiant. To avoid relying on a single random configuration, they repeated the training ten times with different data subsets. Only stars that the network confidently identified as red supergiants in at least eight out of ten runs made it into the final catalog. This weeded out false positives and left 2436 reliable candidates.

When the new red supergiants were plotted on a map, they lined up along the Milky Way's spiral arms. Young, hot OB-type stars were found nearby. This coincidence isn't random: both types are born in the same stellar nurseries — giant gas clouds. So red supergiants can indeed serve as reliable markers of the galaxy's structure. Now astronomers have a tool to refine the shape of the arms and understand what the Milky Way looks like from the outside.