Pulse Shape-Based-Analysis using Machine Learning in the MAJORANA DEMONSTRATOR

 

The MAJORANA DEMONSTRATOR experiment is searching for neutrinoless double-beta decay (0νββ) in ⁷⁶Ge using p-type point contact (PPC) high purity germanium (HPGe) detectors. The data-taking for 0νββ with 30 kg of the enriched detector, 88% in ⁷⁶Ge, has successfully been completed in March 2021. The DEMONSTRATOR continues to operate at the Sanford Underground Research Facility in Lead, SD, with natural HPGe detectors for background studies and other physics. The PPC detector geometry enhances pulse shape analysis to identify different pulses such as multi-site events, surface events, pile-up events, isomeric decay events, and more, and the MAJORANA collaboration has developed traditional algorithms to discriminate events based on their pulse shape. Machine learning approaches using neural networks can also be used for the pulse-shape-based analysis of HPGe detector signals, which is being explored by the collaboration, specifically building interpretable machine learning models. For example, the recurrent neural network model with an attention mechanism allowing it to focus on the rising edge as the most important waveform component gives good performance in identifying the waveform types. This talk will discuss our efforts in developing machine learning approaches to discriminate events with different pulse shapes.