📝 SummarXiv

Abstract

We build a deep neural network model to predict meson widths from quantum numbers and masses based on the Transformer architecture. A Gaussian Monte-Carlo data enhancement method is adopted to enhance the meson data by considering the experimental errors, which significantly increase the data samples and improve the robustness and generalization performance of the model. With the meson widths ranging from $\sim10^{-14}$ to 625 MeV, the relative errors of the predictions behave $0.07\%$, $1.0\%$, and $0.14\%$ in the training set, the test set, and all the data, respectively. The width predictions are presented for the currently discovered mesons and some theoretically predicted states. We also use the model as a probe to study the quantum numbers and inner structures for some undetermined states. Furthermore, this data-driven model is investigated to show well charge conjugation symmetry and approximate isospin symmetry, which is consistent with the physical phenomena. The results indicate that the deep neural network has powerful learning and inference abilities to describe and explore the hadron structures and the complicated interactions in particle physics.