Convention versus Convolution: Two methods of low-energy event classification in CRESST

Abstract

The introduction of new low-threshold detectors in CRESST-III has brought a new phenomenon in the low-energy region to surface. The origin of the Low-Energy Excesshas to date not been identified, with new detectors being crafted in hopes of determining the source of the excess. This work will aim to provide a raw data analysis using conventional CRESST techniques, so-called quality cuts, of event discriminationon one hand, but also build upon the foundation laid in previous works and utilise Machine Learning as a tool for event selection. By employing Convolutional Neural Networks, one of today’s most popular types of Machine Learning algorithms, this work’s analysis will not only be able to reproduce the results achieved with conventional methods, but also allow for a more granular discrimination of event types for low energies.