60,000 artificial signals — that's how many fingerprints it took to train a program to read the patterns left by merging black holes. When two black holes collide, they disturb the very fabric of spacetime: from the site of the black hole merger, gravitational waves (ripples in space) spread out. Detectors like LIGO, built with contributions from Rainer Weiss, pick up these ripples using interferometry (combining light beams) — just as Albert Einstein predicted a century ago.
Each signal is turned into a time-frequency image — a Q-transform. A bright spot means a powerful burst of energy. These images are the true fingerprints of cosmic catastrophes. But here's the catch: two completely different phenomena can produce nearly identical patterns. The first is when the wave, on its way to Earth, bends around a massive object (like a clump of dark matter). This effect, called microlensing (focusing by a compact body), slightly alters the signal's brightness and frequency. The second is when the pair of black holes themselves spin, and their orbit wobbles like a spinning top. This also distorts the pattern, but in a different way.
To quickly distinguish such cases, researchers used Machine Learning. They generated 60,000 numerical simulations of three signal types: normal, distorted by a lens, and distorted by spin. Each signal was run through realistic noise and turned into Q-transforms using signal processing algorithms. These 60,000 images were fed to a Convolutional Neural Network (a program that learns to recognize patterns, like we recognize faces). After training, it learned to determine in seconds what happened to the wave.
The result: on perfect data, accuracy reached 95% — almost flawless. On real LIGO detector noise, it's a bit lower, at 82%, but still impressive. Errors occurred when the distortion was weak or when a signal from black holes with very different masses itself resembled the lensing effect. But crucially, the model confidently distinguishes ordinary mergers from exotic ones. This is a first step toward finding dark matter: if an invisible object distorted the wave, we can spot it. It also prevents errors in measuring distances to black holes. Now every new fingerprint taken from the fabric of the universe can be checked in an instant — bringing us closer to solving the dark side of the cosmos.
A microlens with a mass around a thousand solar masses can alter a merger signal so much that even an automated classifier would notice: the amplitude oscillates at tens of hertz, as if the detector is picking up 'beats' from overlapping ray paths.