The Astronomical Society of the Toms River Area
ASTRA club message board

An international group of astronomers developed a method that allows finding gravitational lenses in giant sets of observational data. This method is based on the same artificial intelligence algorithm that Google, Facebook and Tesla use in recent years. When one galaxy lies behind another galaxy, we sometimes see an obscure galaxy in the form of replicas surrounding the galaxy in the foreground. This phenomenon is called gravitational lensing, because it results from Einstein's General Theory of Relativity, which says that the mass is capable of distorting the trajectory of moving light. Astronomers produce searches for gravitational lenses, as these objects help to better understand the nature of dark matter. "Hunting" for gravitational lenses sometimes requires a lot of patience. Astronomers must review thousands of images. Assistance in this matter is provided by amateur astronomers all over the world. However, recently new telescopes that observe vast areas of the sky have begun to receive more and more pictures, and it has become quite difficult to review all these images manually, even with the help of volunteers.

To solve this problem in a new study, scientists led by Carlo Enrico Petrillo from the University of Groningen, the Netherlands, used an algorithm of computer intelligence called convolutional neural network. Using this algorithm, 761 gravitational lens candidates were detected in a large set of observational data obtained with the Kilo-Degree Survey sky survey. After additional visual control of this set of gravitational lens candidates, astronomers were able to reduce it to 56 objects. These 56 new lenses are still awaiting confirmation with NASA's Hubble Space Observatory (Hubble).

The algorithm for convolutional neural networks was previously used by Google and Facebook to recognize images in photos, and Tesla used it to create unmanned vehicles.

In the future, Petrillo and his colleagues plan to improve their algorithm to ultimately completely eliminate the need for a visual selection phase that requires human participation.

The researchers programmed powerful arrays known as neural networks to serve as a sort of hivelike digital brain in analyzing and interpreting the images of the simulated particle debris left over from the collisions. During this test run the researchers found that the neural networks had up to a 95 percent success rate in recognizing important features in a sampling of about 18,000 images.

It took humanity centuries to decide that Earth orbits the sun. Now a neural network has come to the same conclusion, using the same data, in just a few hours.