Generative adversarial networks for scintillation signal simulation in EXO-200

Generative Adversarial Networks trained on samples ofsimulated or actual events have been proposed as a way of generatinglarge simulated datasets at a reduced computational cost. In thiswork, a novel approach to perform the simulation of photodetectorsignals from the time projection chamber of the EXO-200 experimentis demonstrated. The method is based on a Wasserstein GenerativeAdversarial Network — a deep learning technique allowing forimplicit non-parametric estimation of the population distributionfor a given set of objects. Our network is trained on realcalibration data using raw scintillation waveforms as input. We findthat it is able to produce high-quality simulated waveforms an orderof magnitude faster than the traditional simulation approach and,importantly, generalize from the training sample and discern salienthigh-level features of the data. In particular, the networkcorrectly deduces position dependency of scintillation lightresponse in the detector and correctly recognizes dead photodetectorchannels. The network output is then integrated into the EXO-200analysis framework to show that the standard EXO-200 reconstructionroutine processes the simulated waveforms to produce energydistributions comparable to that of real waveforms. Finally, theremaining discrepancies and potential ways to improve the approachfurther are highlighted.

Journal of Instrumentation, 18, P06005 (2023)