Speaker
Description
In this work, we develop a unified machine learning strategy based on a variational autoencoder inverse mapper (VAIM) to explore the complementarity of the quasi-PDF and pseudo-PDF methodologies. VAIM is a conditional variational autoencoder–based framework that learns a low-dimensional latent representation of nonperturbative QCD information and constructs an inverse map from lattice-measured observables to the underlying partonic structure, enabling a flexible, data-driven reconstruction of PDFs while retaining correlations across different operator insertions and kinematic regimes. Despite a well-grounded theoretical understanding of the interplay between these two approaches, there are currently no established conventions for simultaneously extracting light-cone information from both frameworks in a unified manner. By combining information from the two methods across multiple values of small $z^2$ for the pseudo-PDF method and large $P_z$ for the quasi-PDF method, we aim to investigate how a flexible unified treatment can improve our understanding of higher-twist effects, increase the robustness of lattice determinations of PDFs, and clarify how the quasi-PDF methodology can be more naturally integrated into phenomenological analyses.