Deep Neural Networks for combined neutrino energy estimate with KM3NeT/ORCA6

AbstraktAbstract

KM3NeT/ORCA is a large-volume water-Cherenkov neutrino detector, currently under construction at the bottom of the Mediterranean Sea at a depth of 2450 meters. The main research goal of ORCA is the measurement of the neutrino mass ordering and the atmospheric neutrino oscillation parameters. Additionally, the detector is also sensitive to a wide variety of phenomena including non-standard neutrino interactions, sterile neutrinos, and neutrino decay. This contribution describes the use of a machine learning framework for building Deep Neural Networks (DNN) which combine multiple energy estimates to generate a more precise reconstructed neutrino energy. The model is optimized to improve the oscillation analysis based on a data sample of 433 kton-years of KM3NeT/ORCA with 6 detection units. The performance of the model is evaluated by determining the sensitivity to oscillation parameters in comparison with the standard energy reconstruction method of maximizing a likelihood function. The results show that the DNN is able to provide a better energy estimate with lower bias in the context of oscillation analyses.

KM3NeT/ORCA is a large-volume water-Cherenkov neutrino detector, currently under construction at the bottom of the Mediterranean Sea at a depth of 2450 meters. The main research goal of ORCA is the measurement of the neutrino mass ordering and the atmospheric neutrino oscillation parameters. Additionally, the detector is also sensitive to a wide variety of phenomena including non-standard neutrino interactions, sterile neutrinos, and neutrino decay. This contribution describes the use of a machine learning framework for building Deep Neural Networks (DNN) which combine multiple energy estimates to generate a more precise reconstructed neutrino energy. The model is optimized to improve the oscillation analysis based on a data sample of 433 kton-years of KM3NeT/ORCA with 6 detection units. The performance of the model is evaluated by determining the sensitivity to oscillation parameters in comparison with the standard energy reconstruction method of maximizing a likelihood function. The results show that the DNN is able to provide a better energy estimate with lower bias in the context of oscillation analyses.