Studies of machine learning for event reconstruction in the SNO+ detector and electronic noise removal in p-type point contact high purity germanium detectors

This dissertation presents two distinct topics. Both focus on the development and application of neural networks and deep learning-based methods to rare event searches in physics, specifically neutrinoless double-beta decay. In the first project, a new method for event vertex reconstruction is developed for SNO+ — a large-scale, liquid scintillator-based, multi-purpose neutrino experiment located at SNOLAB in Sudbury, Ontario, Canada. Several studies are conducted to demonstrate its performance in comparison to traditional maximum likelihood reconstruction techniques, as well as its potential to increase the sensitivity of SNO+ to neutrinoless double-beta decay. In the second project, a deep fully convolutional autoencoder is developed and applied to denoise pulses collected from a p-type point contact high purity germanium detector located at Queen's University in Kingston, Ontario, Canada and similar to the germanium detectors used in the arrays of large-scale experiments. It is shown through multiple analyses that denoising using these methods preserves the underlying pulse shape while simultaneously allowing for improvements in the energy resolution and background discrimination power in some circumstances. Detection of the hypothetical neutrinoless double-beta decay could answer long-standing questions in physics and provide a better understanding of the Universe. As such, numerous experiments across the world are running, or under development, to search for this process. While the tools introduced here are applied to a particular liquid scintillator detector and p-type point contact germanium detector, they are broadly applicable to other experimental setups and detection technologies in addition to the specific ones utilized for each project. Furthermore, these tools can be employed to improve the sensitivity of experiments searching for other rare events, such as dark matter, using similar principles. The flexibility and straightforward transfer of these methods are discussed and some ongoing and future work is highlighted. This research is thus relevant both to and beyond the entire rare event search community and has the potential to widely improve analysis techniques, especially in light of the growing size and rates of data collection from modern particle physics experiments.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.