Speaker
Description
Large liquid scintillator detectors are currently searching for neutrinoless double beta decay, a theoretical process that would confirm the existence of Majorana neutrinos. These detectors typically consist of an inner vessel filled with liquid scintillator and surrounded by photomultiplier tubes (PMTs). The production and propagation of photons in the liquid scintillator depend on various optical parameters, such as light yield, attenuation lengths and re-emission/absorption spectra. The different components of liquid scintillator that must be separately characterized result in several dozen independent optical parameters, some of which are highly correlated. This, in addition to the high dimensionality of the data, makes this an ideal problem for machine learning.
The Eos experiment is a 4-tonne fiducial monolithic optical detector. Eos serves as a testbed for next generation detector technologies for neutrino experiments. This work presents the use of simulation-based inference (SBI) to tune the Eos Monte Carlo simulation. Using neural networks we approximate the conditional likelihood as a fast surrogate for iterative simulations. Trained on simulations generated via the RATPAC2 framework, the model learns the relationship between simulated detector events and the simulation parameters such as light yield, scattering length, and absorption length. With the conditional likelihood modeled, Bayesian sampling of the posterior allows us to perform inference on parameter values using extensive calibration data. In the future, we hope to apply this method to larger detectors such as SNO+.