Researchers use machine learning to improve first photo of black hole

Machine Learning

Researchers used machine learning to tighten up previously released images of the black hole. As a result, a portrait of the black hole at the center of the Messier 87 galaxy, which is more than 53 million light-years away from Earth, shows a thinner ring of light and matter surrounding its center. Astrophysics Journal Letter.

The original image was captured by Event Horizon Telescope (EHT) in 2017. This is a network of radio telescopes around the Earth that act as a planet-sized super-imaging tool. The first picture looked like a “fuzzy donut”. NPR, but researchers have used a new method called PRIMO to reconstruct more accurate images. It is a “new dictionary-learning-based algorithm” that learns to “restore high-fidelity images even if they exist.” In other words, using machine learning data based on what we know about the physics of the universe, specifically black holes, to produce better looking and more accurate shots from the raw data taken in 2017. increase.

A black hole is a mysterious and bizarre region of the universe where gravity is so strong that nothing can escape. They form when a dying star collapses under gravity. As a result, the collapse pushes the star’s mass into a smaller space. The boundary between a black hole and its surrounding mass is called the event horizon, and it is the point at which anything that crosses it (light, matter, or Matthew McConaughey) has no return.

“What we’re really doing is learning correlations between different parts of the image. So we’re doing this by analyzing tens of thousands of high-resolution images created from simulations. said Lia Medeiros, an astrophysicist at the Institute for Advanced Study in Princeton, New Jersey, and author of the paper. NPR“If you have an image, pixels close to a given pixel aren’t completely uncorrelated. It’s not like each pixel is doing something completely independent.”

Researchers say the new images match Albert Einstein’s predictions. However, they hope that further research in machine learning and telescope hardware will lead to additional revisions. says Medeiros. “It could be better.”

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *