Speaker
Description
We discuss two applications of machine learning for hadron spectroscopy. The long-debated nature of the Pc(4312) pentaquark candidate has been elucidated using the neural network based classifier. Production of the Pc(4312) state involves two coupled channels p J/ψ and Σ+c D0 thus necessitating the consideration of a 4-sheeted complex energy Riemann surface. Of these four Riemann sheets only 2th and 4th are relevant due to their proximity to the physical region of interest. Combined with two possible physical interpretations considered in our analysis, namely the bound state and virtual state it resulted in four possible class assignments dubbed b|2, v|2, b|4, and v|4. The unitary coupled channel amplitude was considered in the scattering length approximation. By careful statistical analysis, we showed that the v|4 assignment is the most probable. Employing the Principal Component Analysis we showed that just 6 dominant eigenfeatures are sufficient to explain 99% of the signal variance. And projecting the experimental data on these dominant features we showed that they are well represented within the training dataset. Finally, we applied the SHAP (SHapley Additive exPlanations) to identify that near threshold bins are decisive for the class assignment, thus providing an ex-post justification of the scattering length approximation.
A short account of the ongoing joint JPAC-(AI)DAPT analyses to exploit the generative models for data analysis will also be given.
