Speaker
Description
The extraction of Transverse Momentum Dependent distributions (TMDs) from experimental data is a fundamental inverse problem in hadronic physics. Traditionally, this challenge is addressed by assuming specific analytical functional forms. However, these choices can introduce a parameterization bias and limit our ability to explore the full uncertainty of the results. In this talk, we present a novel "Pixel-Based" approach that treats TMD imaging as a discrete reconstruction task. By formulating the TMD convolutions through discrete tensor algebra, we treat each pixel in the impact parameter $b_T$-space as a stochastic variable. This framework allows for a non-parametric reconstruction of the TMDs within a robust Bayesian inference scheme. To efficiently sample the high-dimensional posterior distribution of the pixels, we employ modern Machine Learning techniques, specifically Normalizing Flows. We focus on a systematic study of the resolution limits and stability of this inversion process. By analyzing the resolution matrix in $b_T$-space, we quantify how different kinematic coverages and experimental precision constrain the underlying 3D structure. This work serves as a proof-of-concept for a low-bias imaging framework, establishing a solid foundation for model-independent extractions of nucleon structure.