Researchers used machine learning to improve the first photo of a black hole

The photo algorithm was trained on over 30,000 black hole simulations.

Lia Medeiros / Institute for Advanced Study

Researchers have used machine learning to tighten up a previously released image of a black hole. As a result, the portrait of the black hole at the center of the galaxy Messier 87, over 53 million light years away from Earth, shows a thinner ring of light and matter surrounding its center in a report published today in The Astrophysical Journal Letters.

The original images were captured in 2017 by the Event Horizon Telescope (EHT), a network of radio telescopes around Earth that combine to act as a planet-sized super-imaging tool. The initial picture looked like a “fuzzy donut,” as described by NPR, but researchers used a new method called PRIMO to reconstruct a more accurate image. PRIMO is “a novel dictionary-learning-based algorithm” that learns to “recover high-fidelity images even in the presence of sparse coverage” by training on generated simulations of over 30,000 black holes. In other words, it uses machine learning data based on what we know about the universe’s physical laws — and black holes specifically — to produce a better-looking and more accurate shot from the raw data captured in 2017.

Black holes are mysterious and strange regions of space where gravity is so strong that nothing can escape. They form when dying stars collapse onto themselves under their gravity. As a result, the collapse squeezes the star’s mass into a tiny space. The boundary between the black hole and its surrounding mass is called the event horizon, a point of no return where anything that crosses it (whether light, matter or Matthew McConaughey) won’t be coming back.

“What we really do is we learn the correlations between different parts of the image. And so we do this by analyzing tens of thousands of high-resolution images that are created from simulations,” the astrophysicist and author of the paper Lia Medeiros of the Institute for Advanced Study in Princeton, NJ, told NPR. “If you have an image, the pixels close to any given pixel are not completely uncorrelated. It’s not that each pixel is doing completely independent things.”

The researchers say the new image is consistent with Albert Einstein’s predictions. However, they expect further research in machine learning and telescope hardware to lead to additional revisions. “In 20 years, the image might not be the image I’m showing you today,” said Medeiros. “It might be even better.”