Evaluation of jet energy determination in heavy-ion collisions using machine learning

In high-energy physics experiments, the precise reconstruction of the transverse momentum of jets is challenging. The primary obstacle is the interference caused by background noise particles overlapping with the jet and the limitations of the sensors currently being used. A traditional noise reduction method is area-based subtraction, which calculates the mean value of certain events. In previous studies, some noise reduction methods utilizing machine learning techniques have been proposed. In this study, we evaluate the performance of various machine learning models in reconstructing the jet transverse momentum in simulated heavy-ion events, based on Pythia events embedded in a thermal background distribution. The performance of the models at the particle level is compared to that using pseudo-towers of various resolutions in eta, phi to evaluate the impact of the sensor performance limitations. The results demonstrate that as the resolution decreases, the performance of the machine learning models also degrades, but still performs better than the traditional method. This provides a reference for the application of machine learning methods.