Brookhaven-led AI method improves particle-jet reconstruction in simulations

A Brookhaven National Laboratory-led team developed an AI method that reconstructed particle jets more accurately than conventional methods in simulated heavy-ion collision data. It improved measurements of jets' positions, energies and internal structure, particularly for larger jets where background fluctuations are strongest. The model learns to separate jets from surrounding particles without paired training examples that supply an answer for each crowded detector image. Researchers trained it using simulated collision backgrounds and separate simulated jet images. The work appeared in Scientific Reports. Researchers plan to apply the method to experimental data from sPHENIX, which collected nearly 300 petabytes during its runs; the demonstrated improvement so far comes from simulations.