Over the past few years, cosmologists have been able to place strong constraints on the densities of dark matter and dark energy, despite the fact that their nature is still unknown. One of the researchers' objectives is to establish precise maps of the distribution of dark matter in and between galaxies. In a recent study, cosmologists combined their expertise with that of artificial intelligence experts to develop a neural network capable of estimating, much more precisely than human methods, the quantity and distribution of dark matter.
At ETH Zürich (in Switzerland), physicists and computer scientists have teamed up 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 cosmological data analysis that have much in common with those used for facial recognition by Facebook and other social media. Their findings were recently published in the journal Physical Review D .
While there are no faces to recognize in images of the Universe, cosmologists are always looking for something similar, says Tomasz Kacprzak, a researcher in Alexandre Refregier's group at the Institute of Particle Physics and astrophysics:“Facebook uses its algorithms to find 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 by telescopes, physicists rely on the fact that, like ordinary matter, dark matter slightly bends the path of light rays arriving at Earth from distant galaxies. This effect, known as weak gravitational lensing, very subtly distorts images of these galaxies. Cosmologists can use this distortion to create dark matter distribution maps.
Then they compare these dark matter maps to theoretical predictions to determine which cosmological model best fits the data. This is usually done using human-designed statistics, such as correlation functions, which describe the relationships between different parts of the maps. These statistics, however, are limited in their ability to find complex patterns in maps.
“In our recent work, we used a completely new methodology says Alexandre Refregier. “Instead of inventing the appropriate statistical analysis ourselves, we let computers do the work ". This is where Aurelien Lucchi and his colleagues from the Data Analytics Lab come in. from the computer science department.
They used machine learning algorithms called artificial deep neural networks and taught them how to extract as much information as possible from maps of dark matter. First, computer scientists trained the neural networks by feeding them computer-generated data that simulated the Universe.
Related topic:New map of dark matter distribution suggests revising the speed of evolution of it
This way, they knew what the correct answer should be for a given cosmological parameter (for example, the ratio of the total amount of dark matter to dark energy) for each simulated dark matter map. By repeatedly analyzing dark matter maps, the neural network learned to look for the right kind of features and extract more and more of the desired information.
The results of this training were encouraging:the neural networks produced values 30% more precise than those obtained by traditional methods based on statistical analysis carried out by humans. For cosmologists, this represents considerable progress, since achieving the same precision by increasing the number of telescope images would require twice as much observation time, which is expensive.
Finally, the scientists used their fully trained neural network to analyze actual dark matter maps from the KiDS-450 dataset. “This is the first time that such machine learning tools have been used in this context, and we found that the deep artificial neural network allowed us to extract more data than previous approaches. We believe this use of machine learning in cosmology will have many future applications explains Fluri.
As a next step, he and his colleagues plan to apply their method to larger image sets, such as the Dark Energy Survey . Additionally, more cosmological parameters and refinements, such as details about the nature of dark energy, will feed the neural networks.
