Evaluation of gradient boosting and deep learning algorithms in dimuon production

Abstract

Usage of artificial intelligence (AI) and gradient boosting based machine learning (ML) algorithms in various fields has been favored to advance the human performances due to their accomplishments in real-world issues requiring intensive computation. For this reason, these methods have been started to be applied in physics fields such as material science, particle, and nuclear physics. Dimuons are muon (µ−) and anti-muon (µ+) pairs produced at the different stages of the medium created in high-energy collisions which can be used to understand the formation of the universe. Since they do not interact strongly, they can be utilized to scrutinize properties of the formed medium such as the thermal radiation and production mechanism of various particles. The success of AI and gradient boosting based ML methods suggested their adaptations in particle analysis. In this study, LightGBM, a gradient boosting framework, and Deep Neural Networks (DNNs) performances were examined for the determination of dimuon mass spectrum in proton-proton collisions at the LHC which can be an example for other spectrum studies. From the comparison of the models it is revealed that DNNs performed 99.993% success to determine dimuon invariant mass spectrum with 99.989% sensitivity and 99.989% precision. The results demonsrate the power of nonlinearity property of DNNs to find spectrum in physics studies.