Physicists use machine learning techniques to search for exotic-looking collisions that could reveal new physics

One of the main goals of the LHC experiments is to search for signs of new particles, which can explain many of the unsolved mysteries in physics. Often, searches for new physics are designed to look for one specific type of new particle at a time, using theoretical predictions as a guide. But what about the search for unexpected and unexpected new particles?

Going through the billions of collisions that occur in the LHC experiments without knowing exactly what to look for would be a mammoth task for physicists. So instead of analyzing data and looking for anomalies, the ATLAS and CMS collaborations are letting artificial intelligence (AI) do the work.

At the Rencontres de Moriond conference on March 26, physicists from the CMS collaboration presented the latest results obtained using various machine learning techniques to search for “plane” pairs. These jets are rough sprays of particles that come from strongly interacting quarks and gluons. They are particularly difficult to analyze, but they can hide new physics.

Researchers at ATLAS and CMS use several strategies to train AI algorithms in their aircraft searches. By studying the shape of their complex energy signatures, scientists can determine which particle created the jet.

Using real collision data, physicists in both experiments are training their artificial intelligence to recognize the characteristics of jets coming from known particles. The AI ​​is then able to distinguish between these aircraft and atypical aircraft signatures, which potentially indicate novel interactions. These will appear as an accumulation of atypical aircraft in the data set.

Another method involves instructing the AI ​​algorithm to consider the entire collision event and look for anomalous features in the various detected particles. These anomalous features may indicate the presence of new particles. This technique was demonstrated in a paper released by ATLAS in July 2023, which presented one of the first uses of unsupervised machine learning in an LHC result.

At CMS, a different approach involves physicists creating simulated examples of possible new signals and then tasking the AI ​​with identifying collisions in real data that are different from ordinary aircraft but resemble the simulation.

In recent results presented by CMS, each AI training method showed different sensitivities to different types of new particles, and no single algorithm emerged as the best.

The CMS team was able to constrain the production rate of several different types of particles that produce anomalous jets. They were also able to show that AI-guided algorithms significantly increased sensitivity to a wide range of particle signatures compared to traditional techniques.

These results show how machine learning is revolutionizing the search for new physics. “We already have ideas on how to further improve the algorithms and apply them to different pieces of data to look for certain types of particles,” says Oz Amram, from the CMS analysis team.