CERN uses neural networks to track foreign particles

According to neowin, amedia outlet, research in physics has benefited from the rise of artificial neural networks and deep learning. In the past, we have seen them used to study dark matter and large galaxies. Now scientists have used neural networks in the study of foreign particles. On the compact Proton Collider (CMS) built on the European Nuclear Research Organization (CERN) Large Hadron Collider (LHC), researchers are using neural networks to identify atypical experimental features resulting from proton-proton collisions within the LHC.

CERN uses neural networks to track foreign particles

Traditional collision algorithms have difficulty tracking these experimental characteristics because most of the “fragments” produced by collisions are short-lived. But neural networks can prove effective in this case. This is because they can be trained in actual data.

CMS’s neural networks have been trained with this data and will soon be able to automatically detect experimental features. For training, the researchers used simulation models that improved the probability distribution of jet classes observed in collision data by field adaptation through backward propagation.

Neural networks are trained (under supervision) to distinguish between particle sprays called “jet streams” produced by the decay of long-lived particles and the jet streams produced by more common physical processes.

So far, the model has shown encouraging results. In the process of analyzing particle orbits, there is a 50% chance that the jet stream will be correctly identified from long-life particles, and the model incorrectly identifies the regular jet stream only once per thousand times, showing fewer false positives and false positives.

CERN believes the new system will help advance the agency’s search for transient and singular particles. For more information, you can study the paper published in arXiv.