An Application of Deep Learning in Direct Photon Identification with ALICE’s Forward Calorimeter (FoCal)

The forward electromagnetic and hadronic calorimeter (FoCal) at the ALICE experiment is specifically being developed to measure direct photons in the forward direction, which will enhance the study of the partonic structure of protons and nuclei in the regime of small Bjorken-x. As hadron decays, especially those of neutral pions and eta mesons, can also produce high energy photons which behave as a background to direct photons, the discrimination of direct photons from this background is crucial. To enhance direct photon identification with FOCAL, machine learning is applied. In this contribution, the method of Deep Neural Networks (DNNs) is employed. Not only does DNNs learn the pattern of direct photons and differentiate them from background, it also identifies the input features which are impactful to the particle identification method. The performance of the proposed method is validated using efficiency, purity, confusion matrix, f1-score, and Area Under Curve (AUC). Additionally, it is also compared with the traditional cut-based algorithm