The ARIANNA experiment is an Askaryan detector designed to record radio signals induced by neutrino interactions in the Antarctic ice. Because of the low neutrino flux at high energies, the ability to increase detector sensitivity and data analysis techniques is crucial to maximizing the number of neutrinos measured. In this work, deep learning techniques are explored to improve real-time data collection capabilities and offline neutrino searches. As an introduction, the broader field of multi-messenger astronomy is outlined, an overview of the ARIANNA experiment is provided, and deep learning techniques are detailed. Next, two projects utilizing deep learning to analyze ARIANNA data are presented. In the first project, the amplitudes of the trigger threshold are limited by the rate of triggering on unavoidable thermal noise fluctuations. Here, a real-time thermal noise rejection algorithm is created that enables the trigger thresholds to be lowered, increasing the sensitivity to neutrinos by up to a factor of two (depending on energy) compared to the current ARIANNA capabilities. A deep learning discriminator, based on a Convolutional Neural Network (CNN), is implemented to identify and remove thermal events in real time. This project demonstrated a CNN trained on Monte Carlo data can run on the current ARIANNA microcomputer; the CNN retained 95% of the neutrino signal at a thermal noise rejection factor of 100,000, compared to a template matching procedure which reached only 100 for the same signal efficiency. The results are verified by feeding in generated neutrino-like signal pulses and thermal noise directly into the ARIANNA data acquisition system. There are further studies of the CNN including deep learning network interpretability and hyperparameter optimization. Lastly, the CNN is used to classify cosmic rays events to confirm they are not rejected; the network properly classified 102 out of 104 cosmic ray events as signal.

In the second project, deep learning is used in an offline analysis to classify experimental ARIANNA data collected between 2018-2021. This work compares a more traditional neutrino search technique using cuts on different variables to a new method using deep learning to classify experimental data in an offline analysis. In the second-to-last stage of data cuts, the traditional analysis is found to keep 99% neutrino signal efficiency while rejecting all except 53 experimental background events; the deep learning approach provided significantly better results with 99% signal efficiency while rejecting all except two experimental background events. Both groups of remaining background events were rejected in the final correlation cut stage of the analysis. Due to a limitation in simulating all background event types, the deep learning network was trained on a mixture of simulated and experimental data. A further study was done to check for potential artifacts between the two different types of data that could lead to inaccurate classification results. The study was conducted using the data from a cosmic ray configured ARIANNA station which contained experimentally detected cosmic ray. It is shown through a similar deep learning analysis on the cosmic ray station that there were no artifacts seen in the final model. This provides confirming evidence that artifacts are not affecting the efficiency and background rejection results of the neutrino analysis. This work concludes with a summary of the work done and final recommendations moving forward with deep learning techniques for the ARIANNA experiment.