Machine learning assisted reconstruction of positron-on-target annihilation events in the PADME experiment

Abstract The PADME Experiment at the Laboratori Nationali di Frascati, INFN is used in the search for a Dark photon, produced with an ordinary photon in electron-positron annihilation events. The energy of the photons, emitted in the annihilation is measured using a segmented electromagnetic calorimeter. Machine learning methods consisting of various convolutional neural networks are used for the reconstruction of close-in-time signals with high resolution. These algorithms were used on two-photon annihilation events e + e − → γγ to calibrate the photon energy values. In order to calibrate the neural network output from signal amplitude to energy, the machine learning based results were compared to the conventional methods used for reconstructing the signals. The use of machine learning models for reconstructing real data and the process of calibrating the machine learning method output are presented and discussed.