When two black holes merge, they shake spacetime itself. This ripple, gravitational waves, is detected by giant interferometry laser detectors. But catching the wave is only half the story: we need to know what kind of holes collided. To do that, the waveform is compared with millions of computational templates. Each template is the result of a complex calculation using Einstein field equations, which could take days or weeks.

Researchers have found a way to speed up the process with Machine Learning neural networks. They envisioned a gravitational wave as an orchestra's music. Instead of computing the whole “symphony” at once, they split it into 25 simple pieces — notes for different instruments: rhythm alone, rotation alone, overtones alone. Each note is predicted by a tiny neural network. Then all predictions are assembled — much like a conductor creates a full piece from individual parts. This takes just one millisecond.

Such speed is not just convenient: it changes the game. Processing a single event like GW150914 used to require hours on powerful clusters; now it fits into a few hours on an ordinary graphics card. And the result matches the reference to within a tiny error. Moreover, because the whole model is just a set of formulas and neural networks, it can be tuned, further trained, and easily ported to even more sensitive detectors in the future.