EN | RU

Perfection is achieved, not when there is nothing to add,
and when there is nothing to remove.

Antoine de Saint-exupéry

The use of artificial intelligence will reduce time of processing of astronomical data from weeks and months to seconds

12 Sep 2017

On pages of our site we have repeatedly talked about the effect of gravitational lenses. The effect is that the light from very distant cosmic object, of the galaxy, for example, passing by closer and more massive object massive galaxy or cluster of galaxies. The gravitational forces of the massive object refract and focus light, forming, often, several distorted images of a remote object. The effect of gravitational lenses is one of the most useful tools for astronomers, who are also looking for distant exoplanets, black holes, studying the structure of galaxies and find evidence of the Theory of relativity. However, the analysis of the images obtained by gravitational lenses requires complex mathematical processing and comparing data with mathematical models. And often such analysis requires weeks and months of continuous work is not the weakest of supercomputers.

However, the researchers from the laboratory of linear accelerator SLAC Stanford University have found a way to reduce processing time the images of gravitational lenses to a few seconds. They, within just one day, trained neural network, "feed" her half a million images of the gravitational lens obtained with the help of appropriate mathematical models. After that, the neural network was downloaded for real images obtained with telescopes, and it is in a matter of seconds coped with the task perfectly, restoring the original image of the object in the background with an accuracy that rivals the accuracy when using the methods of traditional treatment.

"The amazing thing in all this is that the neural network was independently trained in the intricacies of identifying all visual features" - the researchers write, "It's very similar to how young children learn to recognize objects in images. You don't have to accurately describe to the child how the dog looks. You just need to show him some relevant pictures. And after some time the child will be able on a picture to identify the breed, size and even the age of the dog."

After the launch of the new space telescope James Webb Space Telescope and the commissioning of new ground-based telescopes, the need for analysis of the images of gravitational lenses will increase many times and the "magic wand" in this case could be the artificial intelligence. Moreover, for the analysis of astronomical data can be used by not the most large and complex neural networks, which can work not only in the depths of the supercomputers for their work will have resources laptop or even a powerful smartphone.