Machine Learning Confirms M87 Black Hole Image, Study Finds

Researchers have independently validated the historic 2019 image of the supermassive black hole at the center of the M87 galaxy using a newly developed machine-learning algorithm.

The study, published on 11 November in Nature, corroborates findings that confirm Albert Einstein’s general theory of relativity.

Machine Learning Confirms Black-Hole Ring

The Event Horizon Telescope (EHT) collaboration revealed the photograph of the supermassive black hole, located 55 million light-years away, on April 10, 2019.

The image shows a bright, lopsided ring of superheated plasma with a dark central shadow. Some researchers argued that the circular shape was an artifact of the team’s image-reconstruction techniques, according to Nature. Starting in 2022, astrophysicist Junichiro Makino of Kobe University and his collaborators questioned those conclusions, with some of their own analyses producing a glob of radio emissions rather than a ring.

To address these discrepancies, a team led by Samuel Lai at Australia’s Commonwealth Scientific and Industrial Research Organisation (CSIRO) applied an independent technique to the original EHT datasets. The researchers used “closure invariants,” which combine triplets of telescope measurements to eliminate atmospheric and instrumental errors, as detailed by Phys.org.

The team developed a machine-learning method named GenDIReCT to further filter out interference and account for biases introduced by combining observations from a sparse global network of telescopes into an Earth-sized virtual interferometer, as reported by Space.com.

An image of the supermassive black hole M87* revealed to the public in 2019
Photo: Space

How the 2019 Image Was Made

The 2019 EHT image resulted from an international collaboration of over 200 astronomers, including scientists from MIT’s Haystack Observatory, who coordinated eight major radio observatories on four continents to function as a virtual, Earth-sized telescope. In a series of papers published in a special issue of Astrophysical Journal Letters, the collaboration revealed four images of the supermassive black hole at the heart of Messier 87, a galaxy within the Virgo galaxy cluster.

Each station recorded data onto several Mark6 units—ultrafast data recorders originally developed at Haystack Observatory—and after the observing run ended, researchers packed up the hard drives and flew them via FedEx to Haystack Observatory in Massachusetts and the Max Planck Institute for Radio Astronomy in Germany.

When the files were released, Akiyama and his colleagues immediately ran the data through their respective algorithms independently to avoid group bias. “This is the dawn of a new era of black hole astrophysics,” said Akiyama.

Australian team confirms first black hol
Photo: UA.NEWS

Training the Model on Everyday Images

To ensure objectivity, the Australian research team trained their artificial intelligence model on everyday terrestrial images such as cats, dogs, and cars. On Dec. 4, 2025, the fully validated imaging method was finally allowed to “look” at the EHT observations of M87, and the model had no idea what a black hole looked like when applied to the data.

Results Match Einstein’s Predictions

The independent reproduction of the M87* ring structure matches the predictions set out in Albert Einstein’s theory of general relativity, published in 1915, which theorized the existence of black holes and the extreme gravitational curvature of spacetime, according to ScienceDaily. The bright, lopsided ring observed by the EHT provides visual confirmation that light bends around the black hole while surrounding material orbits at near-light speed.

“This is the first truly independent validation of the EHT image, an important step in the scientific process,” team leader and researcher with Australia's Commonwealth Scientific and Industrial Research Organisation (CSIRO) Samuel Lai said.

The Event Horizon Telescope project launched in 2009 after decades of theoretical work on photon orbits and technical advances in very-long-baseline interferometry at progressively shorter wavelengths. The collaboration now comprises more than 300 members across 60 institutions in over 20 countries, operating a global array that functions as an Earth-sized virtual telescope by shipping hard drives from sites worldwide to central correlators at MIT Haystack Observatory and the Max Planck Institute for Radio Astronomy.

Machine Learning Confirms M87 Black Hole Image, Study Finds
Photo: NewsBytes

On April 10, 2019, the EHT published the first direct image of a black hole, M87*, in six simultaneous papers in The Astrophysical Journal Letters. Three years later, on May 12, 2022, the collaboration unveiled the first image of Sagittarius A*, the supermassive black hole at the center of the Milky Way, using data collected in 2017 that had been delayed by the seasonal isolation of the South Pole Telescope.

Karl Schwarzschild derived a solution suggesting an infinitesimal region with infinite mass, resulting in unrestrained curvature known as a singularity, surrounded by a spherical boundary—the Schwarzschild radius—at which not even light has the velocity to escape, known as the event horizon.

When scientists first saw the image of M87*, they were astounded at how closely it reflected predictions made by general relativity, thus confirming the theory once again. By reconciling the earlier disagreements regarding the ring’s structure, the study strengthens the scientific consensus on the nature of M87*.