Variable stars pulsate, flare up, or dim due to eclipses — as if performing a cosmic symphony. Their photometry (brightness record) is captured by telescopes with gaps and interference. But this 'sound' holds the secrets of stellar evolution: from birth in gas-and-dust clouds to death as supernovas.

Once, scientists manually searched for signs, now machine learning handles it on its own. The algorithm examines the star in three ways: how its brightness changed, its rhythm (frequencies), and the averaged cycle — like a conductor seeing both the score and the orchestra's performance. This catches even slow shifts invisible to the eye.

The method bypasses manual classification and works well with sparse data: a single example per type is enough. For future surveys counting stars in the billions, this is a must. Unexpectedly, the algorithm proved useful for finding exoplanets — through tiny dips in light. It will also refine cosmic distances via redshift, and together with numerical simulations, predict stellar behavior. Thus, galactic evolution will become clearer as we sort through myriads of light symphonies. Notably: the method's prototype was tested on Mira Ceti, known since antiquity — its discovery a century ago allowed us to gauge the scale of the cosmos.