NeuroMCT: Fast Monte Carlo Tuning with Generative Machine Learning in the JUNO Experiment

The Jiangmen Underground Neutrino Observatory (JUNO) is a neutrino experiment located in China with a wide physics program. The main goals of JUNO are to determine the neutrino mass ordering and to perform highprecision measurements of neutrino oscillation parameters using reactor antineutrinos. To achieve these goals, the JUNO detector comprises an acrylic sphere of 35.4 m in diameter filled with 20 kt of liquid scintillator and is equipped with 43212 photomultiplier tubes (PMTs), providing an energy resolution better than 3% at 1 MeV. A deep understanding of this complex detector is critical for reaching the physics goals. In this regard, a comprehensive and accurate Monte Carlo (MC) simulation of the detector and the physics interactions occurring inside of it is essential. These simulations depend on many different effective parameters that must be tuned or measured to accurately reproduce the acquired data. In this contribution, we propose a machine learning (ML) approach to MC tuning using calibration campaign data. We focus on three key parameters related to the energy response of the JUNO detector: the Birks’ constant k B , the Cherenkov yield factor f C and the light yield Y . We optimize these parameters by comparing calibration data with outputs from ML models, using the Markov chain Monte Carlo method. We study a multi-output regressor and a generative adversarial network as ML methods to rapidly model the calibration sources spectra and as an efficient way to interpolate within the parameter space of ( k B , f C , Y ).