Speaker
Description
The interaction of jets with the color-deconfined QCD medium in relativistic heavy-ion collisions provides crucial insights into the properties of the quark-gluon plasma (QGP). Conventionally, these interactions are studied by measuring modifications in jet observables with respect to proton-proton baselines. Deep learning methods allow us to evaluate the modification of jets on a jet-by-jet basis, significantly improving the capability of using jets to probe the QGP properties.
In this work, we first apply a Dense Neural Network (DNN) approach to reconstruct jet transverse momentum $p_T$ within the complex QGP background, using jets generated with PYTHIA and evolved through the medium with the Linear Boltzmann Transport (LBT) model. We find that the recoil partons in the LBT simulation, scattered out of the QGP background but belonging to jets, can inevitably lead to over-subtraction of the background when the DNN model is trained on PYTHIA jets. By training the DNN model on quenched jets generated with the LBT model, we significantly reduce this prediction bias and achieve more accurate background subtraction compared to conventional area-based and constituent subtraction methods adopted in many experimental measurements.
Going beyond ensemble-averaged jet observables, we further explore the fractional energy loss of each jet through the QGP using the Convolutional Neural Network (CNN) and dynamic graph convolutional neural networks (DGCNNs). The initial jets are generated by PYTHIA, and their subsequent evolution through the QGP is simulated using LBT model that incorporates both elastic and inelastic scatterings between jet partons and the QGP. To emulate realistic experimental conditions, the LBT jets are embedded in a thermal background and subjected to constituent subtraction. We find that, while the CNN performance is degraded by the presence of the background and is not fully recovered after background subtraction, the DGCNN provides more accurate predictions of the jet-by-jet fractional energy loss across the full range of energy-loss values. These results demonstrate the potential of machine-learning-based approaches to improve jet measurements in the QGP and enable more precise jet-by-jet characterization of jet quenching, providing a promising avenue toward jet tomography of the QGP.
[1] Ran Li, Yi-Lun Du, Shanshan Cao, Jet momentum reconstruction in the QGP background with machine learning, Phys. Lett. B 870 (2025) 139940.
[2] Ran Li, Yi-Lun Du, Shanshan Cao, Deep learning for jet modification in the presence of the quark gluon plasma background, Physical Review C 113, 024912 (2026).
