• Physics 16, 63
Using machine learning to fill gaps in telescopic data, researchers have reconstructed a high-fidelity image of the black hole at the center of Messier 87.
In 2019, the Event Horizon Telescope (EHT) collaboration released the first-ever image of a black hole. The image was described by some as a “blurry orange donut”. The EHT includes a global radio telescope, which together create an effective high-resolution, Earth-sized observatory. However, telescopes cannot cover the entire planet, so images must be created from incomplete snapshots from each telescope.A team of researchers now demonstrates the power of machine learning in performing this task [1]Compared to the 2019 image, a new high-fidelity reconstruction of M87 reveals a more defined central region surrounded by a thinner bright ring of accreting gas.
New images are obtained using a machine learning approach called dictionary learning. Dictionary learning uses a large amount of training material to extract rules for analyzing data. This approach is used in image recognition. For example, this kind of algorithm can learn to recognize and analyze images of dogs after being trained on different images of different breeds of dogs, said Dimitrios Psaltis, co-author of the paper. . Here, the researcher created a series of simulated black holes, determined how those black holes would appear in his EHT observations, and used synthetic black hole images to train the algorithm. I created a large suite.
“The most important thing for any training or machine learning algorithm is to make sure the training set has no preconceived notions about what the outcome will be,” says Psaltis. To prevent machine learning algorithms from creating an image of what M87 would look like instead of what a black hole actually looks like, the researchers included an extensive black hole in the training procedure. His 30,000 composite images created by the team were generated from black holes of varying masses and environments with different accretion materials.
Once a machine learning algorithm was trained on these images, the team used it to create images of the black hole from the M87 data collected by EHT. The result was an image with a much lighter orange ring than seen in the original image, and a lighter rim at the bottom.
The researchers found that the resulting images fit well with theoretical expectations. The consistency is “pretty good,” said her Jessica Lu, an astronomy professor at the University of California, Berkeley. He was not involved in this research. “This gives us a lot of confidence that the image they derive is one of the best, if not the best, she can predict from the data,” she says. Lu said the researchers improved her estimate of the size of M87’s emitting ring and used it to derive a more accurate estimate of Black-Her’s hole mass than was possible with older images. He said he is looking forward to it.
“I think it’s amazing what you get by fusing or marrying the world’s best telescopes with modern telescopes. [machine-learning] Algorithms,” says Psaltis. Next, the team plans to apply this algorithm to Sagittarius A*. This is the black hole at the center of our own Milky Way, which he also imaged with EHT (see Research News: First Image of Milky Way Black Hole). Psaltis hopes machine learning algorithms will help researchers dig deeper into their data. The improved image could help us understand, for example, how fast matter moves around the black hole and how much that motion contributes to the blurring of the image. By simply removing the ambiguity caused by data gaps, researchers will be able to glean better information about what’s really happening around black holes, he says.
– Alison Gasparini
Allison Gasparini is a freelance science writer based in Santa Cruz, California.
References
- L. Medeiros and others.“PRIMO Reconstructed Image of M87 Black Hole”, Celestial body. J. , Lett. 947L7 (2023).
