Deep learning for jet modification in the presence of the QGP background

Jet interactions with the color-deconfined QCD medium in relativistic heavy-ion collisions are conventionally assessed by measuring the modification of the distributions of jet observables with respect to their baselines in proton-proton collisions. Deep learning methods enable per-jet evaluation of these modifications, enhancing the use of jets as precision probes of the nuclear medium. In this work, we predict the jet-by-jet fractional energy loss $χ$ for jets evolving through a quark-gluon plasma (QGP) medium using a Linear Boltzmann Transport (LBT) model. To approximate realistic experimental conditions, we embed medium-modified jets in a thermal background and apply Constituent Subtraction for background removal. Two network architectures are studied: convolutional neural networks (CNNs) using jet images, and dynamic graph convolutional neural networks (DGCNNs) using particle clouds. We find that CNNs achieve accurate predictions for background-free jets but degrade in the presence of the QGP background and remain below the background-free baseline even after background subtraction. In contrast, DGCNNs applied to background-subtracted particle clouds maintain high accuracy across the entire $χ$ range, demonstrating the advantage of point-cloud-based graph neural networks that exploit full jet structure under realistic conditions.