Exploring the Performance of Deep Learning in High-Energy Physics

This article presents a comprehensive investigation into the effectiveness of supervised deep learning techniques for classifying the outcome of high-energy particle collisions using CMS Open Data. The research primarily focuses on the conversion of particle and jet position and momentum information into images, followed by the application of convolutional neural networks (CNNs) to classify various particle processes. Two distinct scenarios are considered. The first scenario involves classifying images for processes that generate a known resonance with invariant masses at different energy ranges. The second scenario focuses on identifying signal and background processes with similar final states. Furthermore, alternative CNN architectures are evaluated based on their performance metrics within each scenario. The trained neural network models with the best performance metrics are subsequently employed for classifying real collision data.