I will present my work on the application of generative diffusion models in lattice field theory, a non-perturbative approach to quantum field theory. Diffusion models are a class of generative machine learning models which utilize stochastic processes to draw high-dimensional data from complex probability distributions. While they were introduced mainly for image-generation, they can be applied to a wide range of problems, including lattice field theory. We utilize them to draw field configurations of scalar and U(1) pure gauge lattice field theories. I will discuss the generative performance of the model, the main caveat of this method, and the physical interpretation of the trained neural network. Further, I will talk about the connection between diffusion models and renormalization group transformations.
Thomas et al. (Thu,) studied this question.