MEASUREMENT OF ATMOSPHERIC MUON NEUTRINO DISAPPEARANCE WITH ICECUBE USING CONVOLUTIONAL NEURAL NETWORK RECONSTRUCTIONS

Neutrinos are neutral, fundamental particles that oscillate, or change flavor, while they travel. This phenomenon means that neutrinos deviate from the Standard Model prediction that they are massless, creating an exciting window into physics exploration beyond the Standard Model. Understanding neutrino oscillation and constraining their behavior is crucial to furthering the understanding of these particles and how they fit, or do not fit, into the Standard Model. The IceCube Neutrino Observatory has been detecting neutrinos for more than 10 years, leading to a large sample of atmospheric neutrino data available for studying neutrino oscillations. Reconstructing the neutrino interactions, such as the neutrino’s energy and direction, are key to constraining neutrino oscillation. A fast and robust reconstruction method was developed using convolutional neural networks (CNNs) and optimized to reconstruct parameters necessary to both reconstruct and isolate a pure atmospheric neutrino sample using the IceCube detector. This work compares the performance of this reconstruction to the current likelihood-based reconstruction currently used in IceCube. An analysis of the muon neutrino disappearance is then pursued using 9.28 years of neutrino data. The analysis shows competitive projected sensitivity, the ability to account for numerous systematics, and robust recovery of the physics parameters under statistical and systematic variations. While the 9.28 year sample is still under collaboration review, one year of the total data is used to perform a confirmatory study on the CNN reconstruction and sample. The oscillation parameter constraints from this one year analysis are in alignment with past IceCube analyses and other neutrino experiments within one sigma. This opens the pathway to use the CNN reconstruction for future analyses studying low energy neutrinos on IceCube.