While much of current space news tends to focus on the potential of manned missions within the Earth’s solar system, scientists haven’t lost interest in exploring the deepest reaches of the universe. Though there is a limit to just how far researchers can see, current observational methods still result in massive amounts of potentially useful data. This obviously presents a challenge to scientists, as small teams struggle to sift through this massive store of information. However, European astronomers may have found a solution: artificial intelligence.
Gravitational lenses are a peculiar and rare phenomenon. These rings are created by the light of a distant galaxy being bent by another galaxy closer to the observer. The hoop-like illusion is called an Einstein Ring, since their existence was first suggested by his theory of general relativity. They are a subject of great interest to astronomers researching so-called “dark matter,” a hypothetical category of matter which, while never directly observed, is said to make up about 27 percent of the universe.
Astronomers have historically struggled to find examples of gravitational lenses since it requires sifting through thousands, if not millions of images of a universe with billions upon billions of galaxies, but researchers at colleges and universities in Bonn, Groningen, and Naples are turning to the same artificial intelligence networks used by Tesla, Google, and Facebook to help them in the search. While still in its preliminary stages, the project shows an interesting example of how artificial intelligence and astronomers can work together and allow smaller research teams with limited resources to accomplish more.
The researchers began by creating a database containing millions of homemade images depicting the ever elusive gravitational lenses. These images were then used to train the neural network to identify examples of gravitational lenses from images of small patches of sky, which alone can contain hundreds of thousands of galaxies. In the first major test, the artificial intelligence network was given a small, 255 square degree patch of the night sky, where it identified 761 potential examples of gravitational lenses. The researchers then sifted through the candidates and narrowed the pool down to 56 possible lenses, giving them a new set of objects to investigate and hopefully confirm with Hubble and other powerful telescopes.
While having its 761 objects narrowed down to only 56 might sound unimpressive, using the a.i. is far more efficient than having researchers sift through an ever-growing pool of images. With time, the neural network will only improve in its accuracy, making it a powerful tool with plenty of potential for refinement.