[RBRC seminar] Inferring Parton Distributions: Bayesian & Deep Generative Modeling

by Mr Yamil Cahuana Medrano (College of William & Mary)

Thursday 18 Jun 2026, 12:30 → 14:30 US/Eastern

2-160 (https://bnl.zoomgov.com/j/1600983728?pwd=RAD7OLcqre7Ycsp6JfFp6HAnpyLxex.1)

Extracting parton distribution functions (PDFs) from Lattice QCD is crucial for understanding nucleon structure, but it fundamentally relies on solving a challenging ill-posed inverse problem. In this talk, I will present an overview of my work on solving this problem within the pseudo-PDF framework. First, I will outline how Gaussian processes (GPs) provide flexible Bayesian priors that encode correlations and physical constraints without fixing a functional shape, quantifying information gain through the Kullback–Leibler divergence. Additionally, I will present an alternative non-parametric strategy based on deep generative modeling, utilizing invertible neural networks (INNs) conditioned on a GP prior. Tests on synthetic data confirm the consistency and robustness of both independent methodologies. These results support both GP regression and INN frameworks as systematic approaches to PDF reconstruction, offering controlled uncertainties and reduced model bias in Lattice QCD analyses.