Dark matter: A phenomenon as unknowable as its name suggests, this elusive occurrence is invisible, unidentified and yet somehow ubiquitous, accounting for nearly 90 percent of mass in the universe. Dark matter’s defining properties make it hard to find. However, using object recognition technology, scientists believe they can find identify telltale signs of this scientific mystery.
Astronomers from the universities of Groningen, Naples and Bonn are using artificial intelligence to find a rare phenomenon called a gravitational lens, as reported by EarthSky. First postulated by Einstein, these lenses appear when something with a large amount of mass bends light around it. Such lenses often appear in deep space in the form of a ring of distorted light.
Dark matter is thought to be able to produce such gravitational lenses. Since dark matter is not directly observable, these lenses serve as an important tool for identifying the mass and distribution of dark matter.
However, finding these lenses among countless images of the universe’s 100 billion galaxies can be a daunting task. This magnitude of this problem is only increased by the advent of new telescopes that can capture images of large portions of the cosmos. The number of images now is far too great to review using conventional manual techniques.
To address this challenge, the astronomers trained a convolutional neural network using thousands of images of gravitational lenses. They then fed the network with millions of images taken of a small swath of the heavens, amounting to slightly more than 0.5 percent of the sky.
“Initially, the neural network found 761 gravitational lens candidates,” the scientists stated. “After a visual inspection by the astronomers, the sample was downsized to 56. The 56 new lenses still need to be confirmed by telescopes as the Hubble Space Telescope.”
The astronomers stressed the neural network technology they used is similar to that employed by major technology firms for commercial applications.
“Google employed such neural networks to win a match of Go against the world champion,” the researchers said. “Facebook uses them to recognize what is in the images of your timeline. And Tesla has been developing self-driving cars thanks to neural networks.”
The same technology could be used to find other types of astronomical phenomena.
“This is the first time a convolutional neural network has been used to find peculiar objects in an astronomical survey,” said Carlo Enrico Petrillo of the University of Groningen, first author of the new work. “I think it will become the norm since future astronomical surveys will produce an enormous quantity of data which will be necessary to inspect. We don’t have enough astronomers to cope with this.”
Tyler Schulze is vice president, strategy & development at Veritone. He serves as general manager for developer partnerships, cognitive engine ecosystem, and media ingestion for the Veritone platform. Learn more about our platform and join the Veritone developer ecosystem today.