We believe that dark matter is the unseen force that holds the universe together. It accounts for about 27% of the universe’s contents and about 85% of all matter, but because it is invisible to the naked eye, we must examine its gravitational pull on galaxies and other cosmic structures. One of science’s most perplexing mysteries is still the true nature of dark matter, despite decades of research.
As per a prominent theory, dark matter could be a kind of particle that hardly interacts with anything other than gravity. However, some scientists think that self-interaction, or the occasional interaction of these particles, may occur. Finding such exchanges would provide important hints about the characteristics of dark matter.
It has been extremely difficult to differentiate the faint indications of dark matter self-interactions from other cosmic effects, such as those brought on by active galactic nuclei (AGN), which are supermassive black holes located at the centers of galaxies. It can be challenging to distinguish between the effects of dark matter and AGN feedback because of the similar ways in which they can push matter around.
Astronomer David Harvey at EPFL’s Laboratory of Astrophysics has made a huge advancement by creating a deep-learning algorithm that can separate these intricate signals. Nature Astronomy published the research article.
Their AI-based technique analyzes images of galaxy clusters, which are enormous collections of galaxies held together by gravity, in order to distinguish between the effects of AGN feedback and those of dark matter self-interactions. The breakthrough could significantly improve the accuracy of dark matter research.
Harvey used photos from the BAHAMAS-SIDM project, which models galaxy clusters under various dark matter and AGN feedback scenarios, to train a Convolutional Neural Network (CNN), a type of AI that excels at identifying patterns in images. The CNN was trained to discriminate between signals resulting from AGN feedback and those resulting from dark matter self-interactions by feeding it thousands of simulated images of galaxy clusters.
The most intricate CNN architecture—dubbed “Inception”—turned out to be the most accurate of all the ones tested. Two main dark matter scenarios with varying degrees of self-interaction were used to train the AI, and other models, such as a more complicated, velocity-dependent dark matter model, were used to validate it.
Under optimal conditions, Inception was able to determine with an impressive 80% accuracy whether AGN feedback or self-interacting dark matter influenced galaxy clusters. Even after the researchers added realistic observational noise, which imitates the kind of data we anticipate from upcoming telescopes like Euclid, it continued to perform at a high level.
This implies that Inception, as well as the AI methodology more broadly, may prove immensely helpful for analyzing the vast amounts of data that we gather from space. Furthermore, the AI’s capacity to manage unknown data suggests that it is adaptable and reliable which makes it a promising instrument for upcoming dark matter research.
AI-based methods such as Inception may have a big influence on how we truly comprehend dark matter. This approach will enable scientists to efficiently and precisely sort through the massive amounts of data being collected by new telescopes, possibly providing light on the true nature of dark matter.
