Brookhaven National Laboratory Develops Machine Learning Model for Accurate Identification of Particle Jets in Heavy Ion Collisions

·by Henderson·AI
Brookhaven National Laboratory Develops Machine Learning Model for Accurate Identification of Particle Jets in Heavy Ion Collisions

Physicists at Brookhaven National Laboratory have developed a machine learning model that can automatically identify particle jets from the significant background noise in heavy ion collisions. This advancement will help improve the measurement accuracy of quark-gluon plasma, a state of matter that existed for a microsecond after the Big Bang. The Brookhaven team collaborated with researchers from Columbia University and the University of Colorado Boulder to describe this method in Scientific Reports. The laboratory stated that this is the first time a "unpaired, unsupervised" generative model has been used for this task.

The model learns directly from simulated collision data, rather than relying on labels provided by scientists.

The difficulty in observing jets arises from the fact that in the Relativistic Heavy Ion Collider (RHIC), heavy ions collide at nearly the speed of light, briefly recreating a "soup" of free quarks and gluons. Jets are concentrated bursts of high-energy particles and are one of the best tools for probing this plasma. They lose energy and change structure as they traverse the plasma. However, the signals from jets are obscured by a large background of other particles, a problem that has long limited the accuracy of jet measurements. Brookhaven physicist and Stony Brook University assistant professor Yen-Jie Gao stated, "Improving background reduction can make measurements of interactions between quarks and gluons under extreme conditions more precise."

”

Brookhaven National Laboratory's Machine Learning Model Improves Particle Jet Identification Accuracy

Typically, teaching algorithms requires labeling thousands of examples, such as tagging images as cats or dogs. However, Brookhaven computer scientist Dmitri Tolbuhin pointed out that real experimental data does not provide answer keys, making this approach unsuitable. The new model omits labels. It showcases the pure background of the plasma and ideal simulated jets, deriving how to extract the jet signals while removing the background. Gao likened it to separating a crowded image into two parts: jets and everything else.

The team tested it on simulated data from sPHENIX, a detector that operated during the last three runs of RHIC. In these tests, the model more accurately reconstructed the position, energy, and internal structure of jets compared to traditional background reduction methods, especially for larger jets where background fluctuations are strongest. The current results are based on simulated data.

The next step is to apply it to real data. sPHENIX collected nearly 300PB of data during its run, equivalent to billions of smartphone photos. Gao stated that this will be the first use of such a model in nuclear and particle physics. The team expects it to be applicable for energy measurements at the planned Electron-Ion Collider and the Large Hadron Collider in Europe. Since the model can operate on detector data with similar images, researchers noted it may also assist other experiments that require separating overlapping signals.

This project is related to the Department of Energy's Early Career Award Phase 1, led by Baruch College, with contributions from Brookhaven, Stony Brook University, Columbia University, and the University of Colorado Boulder. Funding from Brookhaven's laboratory-directed research and development, along with support from the Department of Energy's Office of Science, has facilitated this work. Gao praised the close collaboration with Brookhaven computer scientists, stating that this is not a one-time project, and she is excited about the new opportunities for machine learning in science.

H
About the author
Henderson