Can machine learning spot hidden black holes in star clusters?

Researchers trained neural networks on simulations of how black holes grow via collisions in star clusters, then applied these models to real globular and nuclear clusters to predict their central black hole populations. Most globular clusters stay under 100 solar masses, but a few nuclear clusters like NGC 5102 show observed black holes heavier than mergers alone can explain—suggesting gas accretion played a role. The method also quantifies the odds that ancient clusters underwent runaway collisions in their first few million years.