Using Convolutional Neural Networks for Event Pileup Discrimination in CUPID

Neutrinoless double beta decay (0νββ) is a proposed radioactive decay process that could prove the neutrino’s Majorana nature and demonstrate the violation of lepton number conservation. The observation of 0vββ would provide a powerful hint to the origin of the matter-antimatter asymmetry in the Universe. Amongst the next generation of 0vββ experiments, CUPID (CUORE Upgrade with Particle IDentification) will search for this process in the isotope ¹⁰⁰Mo. Since cryogenic calorimeters behave as signal integrators, two events that occur close enough in time result in a total signal that is close to the sum of the two single events. This family of events, called “pileups”, constitutes one of the major sources of background in the region of interest for the CUPID experiment. The pileup of two neutrino double beta decays (2vββ) is expected to be the dominant contribution. Deep learning algorithms trained to identify these anomalies in time series can provide a major improvement to the standard analysis methods. I present an analysis of the efficacy of convolutional neural networks (CNN), a deep-learning architecture typically used for image recognition and processing, as a method of identifying and discriminating pileup events in time series data.