Impact of Machine Learning Tools on Data Analysis for Eos and LiquidO Detectors

As particle detection technology progresses, corresponding improvements in data analysis methods are required to maximize its potential. To address this need, we present three machine-learning-based methods that enhance the precision and speed of event reconstruction, photon extraction, and particle identification in next-generation particle detectors. "HITMAN" uses neural ratio estimators to generate likelihood spaces from forward Monte Carlo simulations. "PLSE" uses a one-dimensional convolutional neural network to extract photon arrival times from digitized waveforms. The GraphNeT project generates graph neural networks that leverage relational information between detector hits.

We implement and evaluate these techniques using the Eos detector and its wide array of calibration sources, which provide a powerful platform for assessing advanced analysis techniques. In Eos, the machine learning approaches are benchmarked against classical data analysis techniques, demonstrating their competitive performance in event reconstruction, pulse extraction, and background rejection. Additionally, we demonstrate the versatility of these methods across detection platforms through application to LiquidO-style detectors, which utilize an opaque scintillator and optical fibers for high-resolution imaging of particle interactions.