AI Is Revolutionizing an Unexpected Field: The Search for New Subatomic Particles

  • The High-Luminosity LHC will be ready in 2030 and produce 40 million collisions per second.

  • AI increases sensitivity to a wide range of particles compared to traditional techniques.

AI revolutionizes an unexpected field: The search for new subatomic particles
No comments Twitter Flipboard E-mail

Artificial intelligence is a powerful tool that physicists are already turning to in their work. In early December 2023, I had the chance to interview Santiago Folgueras, a young particle physicist trained at the University of Oviedo, Purdue University, and the European Organization for Nuclear Research, more commonly known as CERN. Folgueras has won a €1.5 million (roughly $1.6 million) starting grant from the European Research Council to lead a project called "INTREPID."

The ambitious project, which will take at least five years to complete, aims to use artificial intelligence and state-of-the-art programmable cards to improve the filtering system of CERN's Compact Muon Solenoid (CMS) experiment. From a technological standpoint, Folgueras and his team at CERN want to test new cards with built-in artificial intelligence nodes that can perform neural network inference within the card itself. It sounds exciting.

The HL-LHC will produce 40 million collisions per second, generating enormous amounts of information that scientists can't store anywhere. As such, researchers have to develop a system capable of analyzing the data in real-time and making a decision on the collision that just occurred. Ultimately, the system has to decide whether to save or discard the information. That's what INTREPID aims to do.

AI Is Already Making a Difference at CERN

The INTREPID system Folgueras is working on uses AI in real-time to determine whether a collision meets the conditions necessary to be worth saving for later analysis. But beyond this ingenious real-time filtering system, CERN also uses AI for other tasks. The ATLAS experiment and the CMS experiment are already using state-of-the-art machine learning techniques to identify new physics.

Given these circumstances, the researchers have decided to use AI to do much of this analysis and selection work.

Physicists at CERN face an enormous challenge, which is essentially sifting through billions of collisions without knowing what they’re looking for. So far, they’ve tried to identify anomalies, but this is like looking for a needle in a haystack because of the enormous amount of information generated by ATLAS and CMS. Given these circumstances, the researchers have decided to let AI do much of this analysis and selection work.

The experts at CERN have developed several strategies for training AI algorithms. One involves studying the shape of each particle’s energy signature so that the AI can identify which particle it is with a high probability of success. Their algorithm is even able to locate an atypical signature, which has the potential to reveal that a new interaction has occurred.

Another strategy employed by the researchers involves training the AI algorithm to be capable of thoroughly analyzing each collision and identifying whether an anomaly has occurred based on the particles involved in the interaction. Identifying an anomaly could reveal the presence of new particles. These are just two of the strategies CERN physicists are using. The good news is that they’re already showing promising results for one critical reason: AI algorithms significantly increase sensitivity to an extensive range of particle signatures compared to traditional techniques.

Image | S Sioni/CMS-PHO-EVENTS-2021-004-2/M Rayner

More info | CERN

Related | The End of the Universe: What Science Tells Us About the Inevitable Fate of the Cosmos

Home o Index