Simulating neutrino detection events with GANs

This project covers a final-year physics project where light detection events were simulated using generative adversarial networks (GANs).

The research aim was to investigate the ability of deep neural networks to replace Monte Carlo simulations of liquid argon time projection chambers (LArTPCs).

Various architectures of GAN with different input dimensionality were developed and compared, combining different quantities measured in real detectors/simulated using Monte Carlo methods. Novel loss functions were also tested, with Wasserstein GANs chosen as the most feasible way of comparing the deep learning approach to traditional Monte Carlo results.

“