A team of astronomers has made a breakthrough in understanding black holes using artificial intelligence and synthetic data sets. The research, led by Michael Janssen from Radboud University in the Netherlands, involves training a neural network on millions of simulated black hole data sets to extract information that could not be achieved with limited data.
The Event Horizon Telescope (EHT) has released images of supermassive black holes at the centers of galaxies M87 and Sagittarius A* in our own Milky Way. However, the data behind these images contained untapped information that was difficult to analyze.
To overcome this challenge, Janssen’s team used a Bayesian neural network to quantify uncertainties in their analysis. This allowed them to compare EHT data with models more effectively. The researchers found that the black hole at Sagittarius A* is spinning almost at top speed, its rotation axis pointing towards Earth.
The team also discovered that the emission near the black hole was mainly caused by extremely hot electrons in the surrounding accretion disk, rather than a jet. Furthermore, the magnetic fields in the accretion disk behaved differently from theoretical models.
While the results are exciting and challenge prevailing theories, lead researcher Janssen views his approach as a first step. He plans to improve and extend associated models and simulations using data from the Africa Millimeter Telescope when it becomes operational.
The research team’s impressive scaling capabilities were made possible by collaboration with computational services such as CyVerse for data storage and high-throughput computing, OSG OS Pool, Pegasus for workflow management, Germany’s Max Planck Computing and Data Facility, and software tools like TensorFlow, Horovod, and CASA.
Source: https://phys.org/news/2025-06-neural-network-iconic-black-holes.html