Speaker
Description
We present a hybrid computational framework for light-front Hamiltonian calculations in which machine learning is used to assist Fock-space truncation, while the low-lying spectrum is obtained from standard Hamiltonian diagonalization. The goal is to reduce the computational cost of non-perturbative light-front calculations without compromising the physical reliability of established numerical methods. As a proof of concept, we consider simplified light-front bound-state models and train a neural-network-based importance estimator to identify basis states that contribute most significantly to low-lying eigenstates. The selected reduced basis is then passed to a conventional diagonalization procedure, enabling direct comparison with results from larger reference truncations. This approach provides a controlled setting for studying the convergence of light-front spectra under machine-learning-guided basis selection. We discuss quantitative criteria for truncation error, computational efficiency, and the robustness of low-lying wave functions. The framework is intended as a first step toward more efficient light-front calculations relevant to hadron structure and parton physics in the EIC era.