Machine studying might assist reveal undiscovered particles inside information from the Giant Hadron Collider

Scientists used a neural community, a kind of brain-inspired machine studying algorithm, to sift by massive volumes of particle collision information. Particle physicists are tasked with mining this large and rising retailer of collision information for proof of undiscovered particles. Specifically, they’re looking for particles not included within the Commonplace Mannequin of particle physics, our present understanding of the universe’s make-up that scientists suspect is incomplete.

As a part of the ATLAS collaboration, scientists on the U.S. Division of Vitality’s (DOE) Argonne Nationwide Laboratory and their colleagues lately used a machine studying method referred to as anomaly detection to investigate massive volumes of ATLAS information. The tactic has by no means earlier than been utilized to information from a collider experiment. It has the potential to enhance the effectivity of the collaboration’s seek for one thing new. The collaboration includes scientists from 172 analysis organizations.

The group leveraged a brain-inspired kind of machine studying algorithm referred to as a neural community to look the info for irregular options or anomalies. The approach breaks from extra conventional strategies of looking for new physics. It’s impartial of—and subsequently unconstrained by—the preconceptions of scientists.

Historically, ATLAS scientists have relied on theoretical fashions to assist information their experiment and evaluation within the instructions most promising for discovery. This usually includes performing advanced pc simulations to find out how sure features of collision information would look based on the Commonplace Mannequin.

Scientists examine these Commonplace Mannequin predictions to actual information from ATLAS. Additionally they examine them to predictions made by new physics fashions, like these trying to clarify darkish matter and different phenomena unaccounted for by the Commonplace Mannequin.

However up to now, no deviations from the Commonplace Mannequin have been noticed within the billions of billions of collisions recorded at ATLAS. And because the discovery of the Higgs boson in 2012, the ATLAS experiment has but to search out any new particles.

“Anomaly detection is a really totally different means of approaching this search,” stated Sergei Chekanov, a physicist in Argonne’s Excessive Vitality Physics division and a lead creator on the research. “Fairly than searching for very particular deviations, the aim is to search out uncommon signatures within the information which might be utterly unexplored, and that will look totally different from what our theories predict.”

To carry out this kind of evaluation, the scientists represented every particle interplay within the information as a picture that resembles a QR code. Then, the group skilled their neural community by exposing it to 1% of the pictures.

The community consists of round 2 million interconnected nodes, that are analogous to neurons within the mind. With out human steering or intervention, it recognized and remembered correlations between pixels within the photos that characterize Commonplace Mannequin interactions. In different phrases, it realized to acknowledge typical occasions that match inside Commonplace Mannequin predictions.

After coaching, the scientists fed the opposite 99% of the pictures by the neural community to detect any anomalies. When given a picture as enter, the neural community is tasked with recreating the picture utilizing its understanding of the info as a complete.

“If the neural community encounters one thing new or uncommon, it will get confused and has a tough time reconstructing the picture,” stated Chekanov. “If there’s a massive distinction between the enter picture and the output it produces, it lets us know that there may be one thing attention-grabbing to discover in that path.”

Utilizing computational assets at Argonne’s Laboratory Computing Useful resource Middle, the neural community analyzed round 160 million occasions inside LHC Run-2 information collected from 2015 to 2018.

Though the neural community did not discover any obvious indicators of recent physics on this information set, it did spot one anomaly that the scientists suppose is value additional research. An unique particle decay at an power of round 4.8 teraelectronvolts ends in a muon (a kind of elementary particle) and a jet of different particles in a means that doesn’t match with the neural community’s understanding of Commonplace Mannequin interactions.

“We’ll should do extra investigation,” stated Chekanov. “It’s possible a statistical fluctuation, however there’s an opportunity this decay might point out the existence of an undiscovered particle.”

The group plans to use this system to information collected in the course of the LHC Run-3 interval, which started in 2022. ATLAS scientists will proceed to discover the potential of machine studying and anomaly detection as instruments for charting unknown territory in particle physics.

The paper is printed within the journal Bodily Overview Letters.