Speaker
Description
Deep Inelastic Scattering (DIS) at the future Electron-Ion Collider (EIC) will provide high-precision access to the partonic structure of nucleons and nuclei over a wide range of Bjorken-(x) and (Q^2). Therefore, precise knowledge of parton distribution functions (PDFs) and their uncertainties is essential to exploit the full QCD potential of the EIC.
In this work we study an explainable machine-learning framework for the identification and visualization of the regions of ((x,Q^2)) phase space where DIS observables are most sensitive to individual parton flavors and PDF variations. Instead of considering machine learning as a black-box predictor, explainability techniques are employed to construct physics-motivated sensitivity maps which connect the ML response to the underlying partonic structure directly. Such maps can help to identify which EIC kinematic regions are strongest in constraining quark and gluon PDFs.
We have combined explainable ML with DIS kinematics to give observable-level, flavour-dependent sensitivity maps directly relevant to EIC measurements. This provides an intuitive bridge between modern data-driven methods and the traditional QCD interpretation of DIS, with potential applications to both proton and nuclear PDFs.