Dark matter is the invisible force that holds the universe together—or so we think. It makes up about 85% of all matter and about 27% of the universe, but since we can’t see it directly, we need to study its gravitational effects on galaxies and other cosmic structures. Despite decades of research, the exact nature of dark matter remains one of science’s most elusive questions.

One leading theory is that dark matter is a type of particle that barely interacts with anything except gravity. But some scientists believe that these particles can sometimes interact with each other, a phenomenon known as self-interaction. Detecting such interactions would provide important clues about the properties of dark matter.

However, distinguishing subtle signatures of dark matter self-interaction from other cosmic effects, such as those caused by active galactic nuclei (AGN) – supermassive black holes at the centres of galaxies – has been a major challenge. AGN feedback can push matter around in a similar way to dark matter effects, making them difficult to distinguish.

In a major step forward, astronomer David Harvey of the EPFL Laboratory for Astrophysics has developed a deep learning algorithm that can decipher these complex signals. The work was published on: Nature Astronomy.

Their AI-based method is designed to distinguish between the effects of dark matter self-interaction and AGN feedback by analyzing images of galaxy clusters—massive collections of galaxies bound together by gravity. The innovation promises to significantly increase the accuracy of dark matter searches.

Using images from the BAHAMAS-SIDM project, which models galaxy clusters under different dark matter and AGN feedback scenarios, Harvey trained a convolutional neural network (CNN), a type of artificial intelligence that is particularly good at recognizing patterns in images. After taking thousands of simulated images of galaxy clusters, the CNN learned to distinguish between signals resulting from dark matter interacting with itself and signals resulting from AGN feedback.

Of the various CNN architectures tested, the most complex one, called “Inception,” also turned out to be the most accurate. The AI ​​was trained on two basic dark matter scenarios with different levels of self-interaction, and tested on additional models, including a more complex, velocity-dependent dark matter model.

Inception achieved an impressive 80% accuracy under ideal conditions, effectively determining whether galaxy clusters were affected by self-interacting dark matter or AGN feedback. It maintained its high performance even when the researchers introduced realistic observation noise that mimics the data we expect from future telescopes like Euclid.

This means that Inception, and AI approaches in general, could be incredibly useful for analyzing the vast amounts of data we collect from space. The AI’s ability to process unseen data also shows that it is adaptable and robust, making it a promising tool for future dark matter searches.

AI-based approaches like Inception could have a major impact on our understanding of what dark matter actually is. As new telescopes collect unprecedented amounts of data, this method will help scientists sift through it quickly and accurately, potentially revealing the true nature of dark matter.