Benchmarking Computer Learning Characterization of NOvA Test Beam Particle Detections

In previous work, the NOvA collaboration has developed a Convolutional Visual Network (CVN) for the purpose of identifying the species of particles found by NOvA detectors; however, the network’s accuracy has yet to be tested on real detector data. Analysis of the CVN’s performance uses mass calculations and data cuts on other detector-measured variables to determine the species of particles found by the Test Beam detector and compares them to the predictions made by the CVN. Results suggest that the neural network is identifies particles more accurately than the data cuts. Using the output of the CVN to determine further cuts to the data led to an approximately 5-fold increase in agreement between the CVN and data cut methods. Future work should extend methods for finding CVN-determined data cuts for other particle types and investigate if this similarly improves the data cuts’ ability to identify those particles. Improved agreement would provide further evidence for the effectiveness of the CVN model for identifying particles and prove the effectiveness of machine learning in particle physics analysis.