Speaker
Description
An important open question in neutrino physics is the unitarity of the PMNS matrix. Currently, the uncertainties on several matrix elements are too large in order to draw significant conclusions on the unitarity. This is mostly due to the low experimental statistics in the tau neutrino sector.
KM3NeT-ORCA is a water Cherenkov detector under construction with several megatons of instrumented volume. It will observe about 2400 tau neutrinos per year and thus, it will significantly improve the available tau neutrino statistics. In ORCA, tau neutrinos will be identified by observing a statistical excess of cascade-like events with respect to the electron neutrino expectation from the atmosphere. For this purpose, fundamental event properties like the energy and the direction of an event need to be reconstructed based on the experimental data. Additionally, the development of an algorithm for the separation of track-like (mostly $\nu_{\mu}-CC$) and cascade-like (other flavors) neutrino events is necessary. Currently, event properties inspired by the different event types are used with shallow machine learning, in order to discriminate the two classes. Recent advances in computational performance have made it possible to employ deep artificial neural networks. In this approach, the experimental raw data is used for training a deep neural network. Here, the network builds a representation of the typical event properties that can be exploited to distinguish track-like from shower-like events. In this talk, the current status of the ORCA deep learning efforts with respect to the measurement of tau neutrino appearance is presented.
