Machine Learning Analysis for Neutrino Event Reconstruction

Neutrinos are elementary particles that can be created via interactions within cosmic accelerators across the energy spectrum, including within some of the most distant and energetic objects and events in the known Universe. These particles are of particular interest because they only interact weakly and can have extremely high energies, making them unique probes of distant and dense high energy objects. Ultra-high energy (UHE, > 10 PeV) neutrinos can be used as probes to map particle accelerators in space as their weak interaction allows their energy to be unattenuated and their trajectory to be unperturbed on their way to Earth. These particles carry unique information about the places that create them; however, their low cross sections make them incredibly difficult to detect.

The ANtarctic Impulsive Transient Antenna (ANITA) is a balloon-borne experiment that flew above the Antarctic continent four times in the last two decades and is sensitive to the coherent radio frequency (RF) emission from the cascade shower of particles created by a neutrino’s interaction within the Antarctic ice. The emitted RF signal is polarized along the Cherenkov cone of the shower, and this polarization, together with ANITA’s geometry, enables the reconstruction of the neutrino's direction.

With the goal of using neutrinos to extract information about accelerators in space, a central question arises: how accurately can we reconstruct the direction of detected events? This contribution uses publicly-available ANITA simulation code to separately explore whether machine learning techniques can help us to identify patterns we might otherwise miss, possibly improving performance in this area. We develop a Convolutional Neural Network (CNN) to reconstruct the interaction position of neutrinos in the Antarctic ice and compare its accuracy and speed to the existing analysis pipeline. We also develop a modified CNN that incorporates polarization information to reconstruct the original neutrino direction. This approach can be extended and adapted for future neutrino telescopes, such as the Payload for Ultrahigh Energy Observations (PUEO)—ANITA’s predecessor.