AI and the World's Hard Problems · Physics · arXiv · 2026-10-19
Suppression of $^{14}\mathrm{C}$ photon hits in large liquid scintillator detectors via spatiotemporal deep learning
In liquid scintillator neutrino detectors, photons from the beta decay of ¹⁴C contaminate the signal and degrade energy resolution. The authors propose three models, a gated spatiotemporal graph neural network and two Transformer-based models, to tag ¹⁴C photon hits in positron events with ¹⁴C pile-up. On simulated test data they tagged 25–48% of ¹⁴C hits while misidentifying positron hits as ¹⁴C below 1%, which substantially improved total-charge resolution where the two overlap strongly in space and time.