Deep learning reconstruction in ANTARES

TL;DR

This paper explores deep learning techniques to enhance event reconstruction in the ANTARES undersea neutrino telescope, improving directional accuracy and energy estimation to boost sensitivity for neutrino detection and dark matter research.

Contribution

It introduces deep convolutional neural networks for low-energy event direction reconstruction and energy estimation in ANTARES, demonstrating significant performance improvements.

Findings

Improved azimuth angle resolution for low-energy events.

At least doubled sensitivity in low-energy neutrino detection.

Enhanced energy reconstruction accuracy across all neutrino flavors.

Abstract

ANTARES is currently the largest undersea neutrino telescope, located in the Mediterranean Sea and taking data since 2007. It consists of a 3D array of photo sensors, instrumenting about 10Mt of seawater to detect Cherenkov light induced by secondary particles from neutrino interactions. The event reconstruction and background discrimination is challenging and machine-learning techniques are explored to improve the performance. In this contribution, two case studies using deep convolutional neural networks are presented. In the first one, this approach is used to improve the direction reconstruction of low-energy single-line events, for which the reconstruction of the azimuth angle of the incoming neutrino is particularly difficult. We observe a promising improvement in resolution over classical reconstruction techniques and expect to at least double our sensitivity in the low-energy range, important for dark matter searches. The second study employs deep learning to reconstruct the visible energy of neutrino interactions of all flavors and for the multi-line setup of the full detector.