Machine Learning for Ge-Based Neutrinoless Double-Beta Decay Searches

The discovery of the lepton-number-violating neutrinoless double-beta decay would determine the Majorana or Dirac nature of neutrinos, indicate the origin of neutrino mass, and provide a path to leptogenesis in the early universe. ⁷⁶Ge-based searches using High-Purity Germanium (HPGe) detectors have proven to be very successful in searching for this ultra-rare decay, with previous-generation experiments the MAJORANA DEMONSTRATOR (MJD) and GERDA demonstrating the best energy resolution and lowest backgrounds in the field, respectively. This program is rapidly advancing, with LEGEND-200 now taking data, and LEGEND-1000, the proposed the ton-scale phase of the LEGEND program, under design. The low-background requirements of these experiments and well-understood signal formation microphysics in HPGe detectors lead to unique challenges in designing machine learning-based approaches to data analysis, and have led to the development of novel highly-interpretable methods. I'll discuss successful preliminary deployments of machine learning-based analyses in MJD and GERDA, methods that have been developed for LEGEND-200 commissioning and data-taking, and the ongoing research on new methods for LEGEND-1000.