Presentation description

The IceCube observatory is a 1-cubic-kilometer detector located at the geographic South Pole designed to detect neutrinos and cosmic rays. Neutrinos interact with the media and sometimes create secondary relativistic charged particles that produce Cherenkov Light. Digital optical modules (DOMs) observe the light produced by these particles which enables us to reconstruct the properties of the incident neutrino. These neutrinos are detected in three different flavor states, each associated with either electrons, muons, or tau particles, but as they travel through space they oscillate between these states. Atmospheric neutrino oscillation probability is a function of zenith and energy, so precisely reconstructing neutrino interactions inside the detector is critical to physics measurements. Reconstruction for low-energy neutrinos is particularly challenging because less energetic particles hit fewer DOMs, but DeepCore contains DOMs instrumented denser, enabling the detection of lower energy neutrinos. Convolutional Neural Networks (CNNs) have been used to reconstruct the oscillation sample with DeepCore. Using CNNs, I developed a new method to train for direction reconstruction which gives improved zenith resolution and the first low energy azimuth reconstruction with CNNs. In this poster, I show the improved directional reconstruction as well as the improvement on analysis of using the new reconstruction method.