Machine learning may help uncover undiscovered particles in data from the Large Hadron Collider

Scientists used neural networks, a type of brain-inspired machine learning algorithm, to sift through large amounts of particle collision data. Particle physicists are tasked with mining this vast and ever-growing amount of collision data for evidence of undiscovered particles. In particular, they are looking for particles that are not included in the Standard Model of particle physics, the current understanding of the composition of the universe that scientists suspect is incomplete.

As part of the ATLAS collaboration, scientists at the U.S. Department of Energy (DOE) Argonne National Laboratory and their colleagues recently analyzed large amounts of ATLAS data using a machine learning approach called anomaly detection. This method has never been applied to data from collider experiments before. It has the potential to improve the efficiency of collaboration in the search for something new. This collaboration involves scientists from 172 institutions.

The researchers used a type of brain-inspired machine learning algorithm called a neural network to search the data for unusual features and anomalies. This technique breaks away from traditional methods of exploring new physics. It is independent of and therefore unconstrained by the scientists' preconceptions.

Traditionally, ATLAS scientists have relied on theoretical models to guide experiments and analyzes in the directions most promising for discovery. This often involves running complex computer simulations to determine what certain aspects of crash data look like according to standard models.

Scientists compare these standard model predictions with actual data from ATLAS. They also compare them to predictions made by new physical models that try to explain dark matter and other phenomena that cannot be explained by the standard model.

But so far, no deviations from the standard model have been observed in the billions of collisions recorded by ATLAS. And since the Higgs boson was discovered in 2012, the ATLAS experiment has yet to discover any new particles.

“Anomaly detection is a completely different way to approach this exploration,” said Sergey Chekanov, a physicist in Argonne University's High Energy Physics Department and lead author of the study. “Rather than looking for very specific deviations, the goal is to find unusual signatures in the data that have not been fully explored and may be different from what our theory predicts.”

To perform this type of analysis, the scientists represented each particle interaction in the data as an image similar to a QR code. The team then trained a neural network by exposing it to 1% of his images.

This network is made up of approximately 2 million interconnected nodes, similar to neurons in the brain. We identified and memorized correlations between pixels in images that characterize standard model interactions without human guidance or intervention. In other words, it has learned to recognize typical events that fit the standard model's predictions.

After training, the scientists fed the remaining 99% of images through the neural network to detect anomalies. Given an image as input, a neural network is tasked with recreating the image using its understanding of the entire data.

“When a neural network encounters something new or unusual, it gets confused and has a hard time reconstructing the image,” Chekanov says. “If there's a big difference between the input image and the output it produces, you know there might be something interesting to explore in that direction.”

Using computational resources at Argonne's Laboratory Computing Resource Center, the neural network analyzed approximately 160 million events in LHC Run-2 data collected from 2015 to 2018.

While the neural network didn't find any obvious signs of new physics in this dataset, it did find one anomaly that scientists think is worth further study. The exotic particles decay at energies of about 4.8 teraelectronvolts, producing jets of muons (a type of fundamental particle) and other particles in a way that is inconsistent with our understanding of standard model interactions in neural networks.

“Further investigation is needed,” Chekanov said. “While this could be a statistical fluctuation, this decay could indicate the presence of undiscovered particles.”

The team plans to apply the technique to data collected during LHC Run-3, which began in 2022. ATLAS scientists will continue to explore the potential of machine learning and anomaly detection as tools to unlock the unknowns of particle physics.

The paper will be published in a magazine physical review letter.