Voir en


At CMS, artificial neural networks search for exotic particles

Machine-learning technology is being used by the CMS collaboration to track odd events among LHC data

CMS Event Displays for EXO-19-011
A simulated CMS collision where a long-lived particle is produced together with other 'regular' jets. The long-lived particle travels for a short distance before it decays, creating particles that appear displaced from the point where the LHC beams collided (Image: CMS/CERN)

The CMS Collaboration has developed an artificial neural network that can identify exotic particles generated by the proton–proton collisions inside the LHC.

The “long-lived” particles chased after by the experiments can travel measurable distances (fractions of millimetres or more) from the collision point inside each LHC experiment before decaying. Most of these long-lived particles are undetectable, but could decay to detectable particles, which would lead to a rather atypical experimental signature.

This is where CMS’s new tool comes into play. Standard algorithms used to interpret the data from proton–proton collisions are not designed to seek out such odd-looking events. Artificial neural networks, though, can automatically learn from data to achieve their goal. CMS’s neural network has been fed data from real collision events to train it to spot the odd events by itself.

This project is part of a larger, coordinated effort across all the LHC experiments to use modern machine-learning techniques that will improve the recording and analysis of the large amounts of LHC data. It marks a new step forward after decades of artificial-intelligence use in the field of high-energy particle physics.

Read the full story on the CMS website.