Deep Learning Applications for Neutrino Event Reconstruction in Liquid Argon Time Projection Chamber Detectors and Measurement of Zero-Pion Production

The understanding of several aspects of neutrino interaction physics is still driven by small samples of just a few hundred events recorded in the bubble chamber era in the 70's and 80's, and by a small number of modern, high-statistics but low-resolution measurements. Liquid-Argon (LAr) Time Projection Chamber (TPC) detectors that are now coming online offer imaging capability comparable to that of bubble chamber experiments but much higher statistics and they will bring a generational advance in neutrino studies. By the end of this decade, some of most precise measurements of neutrino interaction characteristics will come from the Fermilab short-baseline (SBN) programme and, in particular, from the SBND experiment starting data-taking operation in late 2020.

The goal of this project is to perform precision measurements of exclusive CC interaction channels with 0-pions and with 0, 1, 2 or more protons in the final state. This is a key set of measurements that will enable us to map the nuclear response function, to disentangle genuine (bare) CCQE interactions from multi-nucleon interactions and inelastic backgrounds, and to improve the systematics of the physics models that will inform the CP searches in the early exploitation phase of DUNE.
The accurate reconstruction of neutrino interactions and, in particular the removal of cosmics (a real challenge for a LArTPC detector operating on surface), the reconstruction of the interaction vertex, and the identification and reconstruction of very short proton tracks near the interaction vertex will be crucially important for this project. An important part of this project will be the development of new neutrino event reconstruction methods based on Deep Learning approaches, the validation of these methods against SBND data, and their deployment in DUNE.