Novel Calibration and Inference Techniques for a Liquid Xenon Dark Matter Detector

XENONnT is one of the leading dark matter detectors aiming to unveil the mystery of Weakly Interacting Massive Particles (WIMPs). As the detector sensitivity improves, we are approaching the threshold for detecting solar neutrinos via coherent elastic neutrino-nucleus scattering (CEvNS). Since these solar neutrinos deposit only O(1keV) of energy in the detector, it is crucial to understand the detector's response to these near-threshold energy depositions. In this thesis, a novel neutron calibration for XENONnT using a ⁸⁸YBe photoneutron source is comprehensively presented to study the detector's near-threshold response. In particular, machine learning methods were applied to enhance signal-background discrimination in the analysis of the calibration. This calibration enables accurate measurement of the solar ⁸B neutrino CEvNS and the search for light dark matter particles. In recent years, the concept of simulation-based inference has gained popularity, as machine learning models offer powerful capabilities for generating probability distributions. This thesis also explores the application of flow-based models in WIMP search inference and compares their performance with traditional methods. The strong agreement between the results obtained from both approaches validates the feasibility of using simulation-based inference in WIMP searches. As simulation-based inference can naturally and flexibly model complex, high-dimensional distributions, it holds great promise for future dark matter searches.