Training Convolutional Neural Networks with Artificial Data to Classify Scintillator Data

Isomeric states are sensitive to changes in nuclear structure. Observing evidence of isomeric transitions is important to expanding our knowledge of nuclear structure as a whole. Our detector is a monolithic inorganic planar scintillator coupled to a position-sensitive photomultiplier tube, designed to measure energy deposited from β particles, internal conversion electrons, and γ rays. However, energy depositions that occur very close together in space and/or time are unrecoverable by the current analysis pipeline, and the datasets' size makes hand-analysis impossible. This drives the development of a new method to produce artificial scintillator data and a machine learning solution based on convolutional neural networks trained on artificial data for the purpose of classifying experimental data. This method has promising results in both classification of single vs multiple interactions and using regression to find the energy and position of interactions in artificial scintillator data.