The telescope of the Vera Rubin Observatory will send millions of flares every night — spotting a rare one among them is like hearing a whisper in the roar of the surf. To train the machine learning algorithm, scientists created thousands of fakes: using physics, they modeled light curves for 14 types of events, from supernovae (star explosions; term coined by Zwicky, mass limit by Chandrasekhar), kilonovae and active galactic nuclei to stars torn apart by black holes, adding time dilation and dimming by cosmic dust.

The system assesses brightness changes in different colors, accounts for light reddening with distance, and delivers a verdict in a fraction of a second. Accuracy reached 80%. This makes it possible not to miss fleeting kilonovae — sources of heavy elements and gravitational waves.