Understanding Backgrounds Using Deep Learning at the Daya Bay Experiment

The Daya Bay experiment uses reactor antineutrino disappearance to measure the neutrino mixing parameter θ₁₃. A variety of deep neural networks are tested with a well-understood uncorrelated accidental background to the inverse beta decay signal to assess the utility of deep learning approaches for characterizing and discriminating backgrounds. Crucially, the training procedures are data-driven and do not rely on simulated events to train the neural networks. The eventual goal of this technique is to reduce the correlated β-n background, which results from the decay of ⁹Li produced by cosmic-ray muons. This background contributes the largest systematic uncertainty in the determination of θ₁₃.