Credit: ETH Zurich
Understanding how our universe has become what it is today and what its final destiny will be is one of the greatest challenges of science. The impressive display of countless stars on a clear night gives us an idea of the extent of the problem, yet this is only part of the story. The deepest enigma lies in what we cannot see, at least not directly: dark matter and dark energy. With the dark matter that unites the universe and the dark energy that makes it expand faster, cosmologists need to know exactly how much of those two is out there to perfect their models.
At the ETH in Zurich, scientists from the Department of Physics and the Department of Computer Science have now joined forces to improve standard methods for estimating the dark matter content of the universe through artificial intelligence. They used state-of-the-art machine learning algorithms for analyzing cosmological data that have much in common with those used for facial recognition by Facebook and other social media. Their results have recently been published in the scientific journal Physical revision D.
Facial recognition for cosmology
While there are no faces to recognize in the photos of the night sky, cosmologists are still looking for something similar, as explained by Tomasz Kacprzak, researcher of the group of Alexandre Refregier at the Institute of particle physics and astrophysics: "Facebook uses its own algorithms for finding eyes, mouths or ears in images, we use ours to look for the telltale signs of dark matter and dark energy ". Since dark matter cannot be seen directly in the telescope images, physicists rely on the fact that all matter, including the dark variety, slightly bends the path of light rays arriving on Earth from distant galaxies. This effect, known as "weak gravitational lens", very subtly distorts the images of those galaxies, just as distant objects appear blurred on a hot day while light passes through layers of air at different temperatures.
Cosmologists can use that distortion to work backwards and create mass maps of the sky that show where dark matter is. Next, they compare those dark matter maps with theoretical predictions in order to find which cosmological model best matches the data. Traditionally, this is done using statistics designed by man as the so-called correlation functions that describe how the different parts of the maps are related to each other. These statistics, however, are limited on how they can find complex patterns in matter maps.
Neural networks teach themselves
"In our recent work, we used a completely new methodology," says Alexandre Refregier. "Instead of inventing the appropriate statistical analyzes ourselves, we let computers do their job." This is where Aurelien Lucchi and her colleagues from the Data Analytics Lab of the Computer Science Department come. Together with Janis Fluri, a doctoral student in the Refregier group and lead author of the study, they used machine learning algorithms called deep artificial neural networks and taught them to extract as much information as possible from dark matter maps.
Related article: Are we looking for dark matter in the right places?
At first, scientists trained neural networks by providing them with computer-generated data that simulates the universe. In this way, they knew what the correct answer would be for a given cosmological parameter, for example the relationship between the total amount of dark matter and dark energy, for each map of the simulated dark matter. By repeatedly analyzing the maps of dark matter, the neural network has taught itself to look for the right kind of features in them and to extract more and more desired information. In Facebook's analogy, it has improved in distinguishing random oval shapes from the eyes or from the mouth.
More accurate than the analysis made by the man
The results of that training were encouraging: neural networks presented values that were 30 percent more accurate than those obtained by traditional methods based on statistical analysis made by man. For cosmologists, this is a huge improvement as achieving the same accuracy by increasing the number of telescope images would require twice the observation time, which is expensive.
Finally, the scientists used their fully trained neural network to analyze the actual dark matter maps from the KiDS-450 data set. "This is the first time such machine learning tools have been used in this context," says Fluri, "and we have discovered that the deep artificial neural network allows us to extract more data from the data than the previous approaches. We believe this use of machine learning in cosmology will have many future applications ".
As a next step, he and his colleagues plan to apply their method to larger image sets like the Dark Energy Survey. Furthermore, more parameters and cosmological refinements such as details on the nature of dark energy will be fed to neural networks.
Do you like this article? Click here to subscribe to free newsletters from Lab Manager