An Update in Deep Neural Network Extraction of Compton Form Factors from DVCS Observables

Understanding the internal structure of the nucleon remains a central challenge in modern nuclear physics. One approach to probing this structure is through the study of deeply-virtual Compton scattering (DVCS), which provides access to Generalized Parton Distributions (GPDs). These GPDs, however, do not directly enter into observables; Compton Form Factors (CFFs), on the other hand, do. These structure functions encapsulate the nonperturbative aspects of nucleon structure. In this work, we investigate a data-driven framework for the extraction of all eight CFF components (real and imaginary parts s) living in the DVCS process using deep neural networks (DNNs). The models are trained and validated on both pseudodata (generated from phenomenological GPD models such as KM15) and real DVCS measurements from experiments. This approach aims to assess the consistency, stability, and interpretability of CFF extraction under realistic experimental uncertainties. The long-term objective is to develop a globally consistent, model-independent representation of the CFFs across kinematic regimes, paving the way for direct comparison with theoretical GPD parametrizations. We discuss progress in local fitting of various observables (thereby CFFs) under simplified (kinematically-justifiable) situations at fixed kinematic settings in both pseuodata and experimental contexts to assess an incremental approach to the extraction. We also discuss the attempt of a "simultaneous fit," so-called because we obtain local fits of CFFs through the use of more than one DVCS observable in the DNN objective function. Finally, we will discuss preliminary results involving the "stitching together" of local fits across a broad range of kinematic settings to approach the extraction of a global model of the CFFs.