CSIRO Researchers Verify First Black Hole Image With Machine Learning

Researchers have independently verified the first-ever image of a supermassive black hole, confirming the 2019 findings that provided a visual cornerstone for Albert Einstein’s theory of general relativity.

The validation arrives from a team at Australia’s Commonwealth Scientific and Industrial Research Organisation (CSIRO), which published its findings Nov. 11 in the journal Nature. By using a machine learning technique called GenDIReCT, the team reproduced the ring-like structure of M87*—the black hole located 55 million light-years away—without relying on the original imaging algorithms.

Independent Verification of M87*

When the Event Horizon Telescope (EHT) collaboration first released the image of M87*, it represented a historic milestone. However, the process of reconstructing images from sparse telescope data left room for debate. According to the team’s research paper, various imaging heuristics led to divergent representations of fine-scale features, with some external analyses even questioning the fundamental ring-like morphology of the black hole.

“It is like standing in your backyard and trying to take a photo of a doughnut on the moon,” Samuel Lai, a researcher with CSIRO, told Space.com.

To overcome these challenges, the team designed their model to filter out interference from Earth’s atmosphere and biases inherent in combining multiple telescopes into an Earth-sized interferometer. The original image, which depicts a supermassive black hole with a mass roughly 6.5 billion times that of the sun, served as a powerful confirmation of general relativity.

Published in 1915, Einstein’s theory predicted that mass curves spacetime, creating a gravitational boundary known as the event horizon where light cannot escape. The image’s ring of superheated plasma, cast against the shadow of the black hole, perfectly matched those long-standing predictions.

CSIRO Researchers Verify First Black Hole Image With Machine Learning
Photo: NewsBytes

“When we applied it to the original EHT dataset, we found a black hole almost identical in size and shape as the historic EHT image,” Lai told Space.com.

“Importantly, our model had no idea what a black hole looked like when we applied it to the M87 data,” Lai said.

Einstein’s Black Hole Prediction Gains Independent Support
Photo: JournalArta

Expanding the Scientific Toolkit

The EHT collaboration has welcomed the independent check. Laurent Loinard, Director of the EHT Collaboration, said the GenDIReCT method successfully recovered the ring structure across several years of data.

“Studies like this strengthen our confidence in what we have learned about black holes, and enrich the toolkit available to the whole community,” Loinard said.

The researchers suggest that their machine learning approach could eventually extend beyond astrophysics, with potential applications in medical imaging and the mapping of earthquakes. While the 2019 image was widely celebrated, the complex process of reconstructing telescope data using different algorithms produced varying results. This new study provides an independent validation that confirms the original ring-like structure without using the same prescriptive morphological priors.

CSIRO, the national science agency that produced the GenDIReCT verification, traces its lineage to the Advisory Council of Science and Industry established in 1916. It became the Council for Scientific and Industrial Research in 1926 and adopted its current name in 1949.

From its Canberra headquarters it now operates more than 50 sites in Australia, France and the United States with 6,618 staff. Its Space and Astronomy business unit manages the Australia Telescope National Facility, which includes the Parkes Observatory and the Australian Square Kilometre Array Pathfinder.

GenDIReCT is a machine learning technique developed by the CSIRO team to process interferometric data. It accounts for atmospheric interference and the specific biases introduced by combining observations from a global network of telescopes.

By confirming the ring structure independently, the study removes doubts regarding whether atmospheric conditions or image-processing choices influenced the initial discovery. The study, titled “Independent interferometric observable confirms black hole shadow in M87*,” is available in the journal Nature.