Deep Neural Networks for Simulation, Reconstruction, and Analysis of EXO-200 data

EXO-200 was an experiment whose primary goal was to search for the neutrinoless double-beta decay of Xe-136. During its lifetime, the experiment accumulated unique datasets of physics and calibration data that remain to be a perfect testbed for advanced analysis techniques, such as deep neural networks (DNNs). After introducing the EXO-200 experiment, this talk describes successful applications of DNNs to simulate, reconstruct, and analyze the EXO-200 data. We show that a DNN is able to extract necessary high-level information directly from raw digitized waveforms, with minimal pre-processing. The accuracy of the developed algorithms is presented and compared to what was achieved by the conventional approaches. Importantly, the methods described in this talk are validated by (and in some cases are trained on) the real detector data, either reducing or eliminating the reliance on the Monte Carlo and its imperfections. The talk concludes by discussing advantages and challenges of DNNs for the upcoming low-background experiments, in particular nEXO and DARWIN.