Signal-background Discrimination in EXO-200 with Deep Learning

We develop a deep neural network (DNN) algorithm to discriminate signal and background events in the EXO-200 neutrinoless double beta decay experiment. In the studied case, the DNN is able to classify Monte Carlo events as signal (neutrinoless double beta decay of ¹³⁶Xe) or different backgrounds directly from the waveforms. The classification accuracy exceeds that of conventional algorithms used in the experiment, trained over the same data. The accuracy of background classification is checked with real detector calibration data for the most important background types (gamma decays of ²³⁸U/²³²Th decay chains and ⁶⁰Co, two-neutrino double beta decay of ¹³⁶Xe, beta decay of ¹³⁷Xe). Additional checks to substantiate the validity of the network’s performance will be presented.