Identification of new physics signals based on deep learning techniques in dark matter direct detection experiments

Abstract

The constant effort to develop better detectors in the search of Dark Matter (DM) has made them more sensitive to noise in the measurements making more difficult the task of event classification were liquid xenon detectors arise as a promising way of finding DM. Leveraging advancements in Deep Learning (DL), this work presents a novel framework for classifying events from DM detectors under the Standard Model of Particle Physics (SM) and Beyond Standard Model (BSM) neutrino interactions. The proposed methodology combines simulations of detector interactions and Neural Network (NN) based classification to tackle the challenges posed by degeneracy caused by overlapping signals. This framework was tested using simulated data presenting degeneracy with the SM neutrinos considering Weakly Interacting Massive Particle (WIMP)s and neutrino vectorial Non-Standard Interactions (NSI), yielding high classification performance with mean Area Under the receiver operating characteristic Curve (AUC) scores approaching 1. Key contributions include the development of a novel event classification framework and insights into the parameter space regions of WIMPs and BSM neutrinos. These findings not only validate the feasibility of applying the Deep Support Vector Data Description (Deep-SVDD) model but also demonstrate its potential for a more detailed framework and its applications in the fields of DM and particle physics.