An international team of astronomers has trained a neural network with millions of synthetic simulations and artificial intelligence (AI) to uncover new cosmic mysteries about black holes, revealing that the one at the center of our Milky Way is spinning at almost top speed, according to a study published in Astronomy & Astrophysics.
Artist impression of a neural network that connects the observations (left) to the models (right). Image Credit: EHT Collaboration/Janssen et al.
The Center for High Throughput Computing (CHTC), a collaboration between the University of Wisconsin-Madison and the Morgridge Institute for Research, supplied the throughput computing resources that produced these enormous ensembles of simulations.
This year marks the 40th anniversary of high-throughput computing, which was invented by Wisconsin computer scientist Miron Livny. By automating computations across a network of thousands of machines, this innovative type of distributed computing effectively transforms a single, enormous computing problem into a supercharged fleet of smaller ones.
This computing innovation is accelerating big-data discovery in hundreds of scientific initiatives across the world, including the search for cosmic neutrinos, subatomic particles, and gravitational waves, as well as the unraveling of antibiotic resistance.
In 2019, the Event Horizon Telescope (EHT) Collaboration revealed the first image of a supermassive black hole in the heart of the galaxy M87. In 2022, they revealed the picture of Sagittarius A, the black hole at the heart of the Milky Way galaxy. However, the data behind the photos still held a lot of difficult-to-understand information. A multinational team of academics developed a neural network to extract as much information as possible from the data.
From a Handful to Millions
Previously, the EHT Collaboration employed just a few realistic synthetic data sets. The National Science Foundation (NSF) funded the Madison-based CHTC as part of the Partnership to Advance Throughput Computing (PATh) initiative, allowing astronomers to feed millions of such data files into a Bayesian neural network, which can quantify uncertainty. This enabled the researchers to perform a far more accurate comparison of the EHT data and models.
Thanks to the neural network, the researchers now believe that the black hole at the heart of the Milky Way is spinning at nearly peak speed. Its rotating axis points to the Earth. Furthermore, the emission near the black hole is mostly created by extremely hot electrons in the surrounding accretion disk, not a so-called jet. Furthermore, the magnetic fields in the accretion disk appear to behave differently than in previous ideas of such disks.
That we are defying the prevailing theory is of course exciting. However, I see our AI and machine learning approach primarily as a first step. Next, we will improve and extend the associated models and simulations.
Michael Janssen, Study Lead Researcher, Radboud University Nijmegen
Impressive Scaling
The ability to scale up to the millions of synthetic data files required to train the model is an impressive achievement. It requires dependable workflow automation, and effective workload distribution across storage resources and processing capacity.
Chi-kwan Chan, Associate Astronomer and PATh collaborator, Steward Observatory, University of Arizona
“We are pleased to see EHT leveraging our throughput computing capabilities to bring the power of AI to their science. Like in the case of other science domains, CHTC’s capabilities allowed EHT researchers to assemble the quantity and quality of AI-ready data needed to train effective models that facilitate scientific discovery,” said Anthony Gitter, a Morgridge Investigator, Professor, and a PATh Co-PI.
The NSF-funded Open Science Pool, managed by PATh, provides computer capacity supplied by over 80 universities across the United States. Over the past three years, the Event Horizon black hole project has carried out nearly 12 million computing jobs.
A workload that consists of millions of simulations is a perfect match for our throughput-oriented capabilities that were developed and refined over four decades. We love to collaborate with researchers who have workloads that challenge the scalability of our services.
Miron Livny, Director and Path Lead Investigator, Center for High Throughput Computing, Morgridge Institute for Research
Journal References:
Janssen, M., et al. (2025) Deep learning inference with the Event Horizon Telescope I. Calibration improvements and a comprehensive synthetic data library. Astronomy & Astrophysics.doi.org/10.1051/0004-6361/202553784
Janssen, M., et al. (2025) Deep learning inference with the Event Horizon Telescope II. The Zingularity framework for Bayesian artificial neural networks. Astronomy & Astrophysics. doi.org/10.1051/0004-6361/202553785
Janssen, M., et al. (2025) Deep learning inference with the Event Horizon Telescope III. ZINGULARITY results from the 2017 observations and predictions for future array expansions. doi.org/10.1051/0004-6361/202553786