Development of a Diffusion Model for Stochastic High Energy Particle Event Generation

Monte Carlo (MC) simulations of particle showers from nuclear collisions are an important tool for analysis of data from nuclear physics experiments. The accuracy of these simulations is limited due in part to the imperfect modeling of particle reaction dynamics in the detector material but also due to the large number of experimental factors that influence the response of the detector to incident particles. With the advancement of machine learning and artificial intelligence has come a promising means to address these shortcomings. Unlike MC simulations, machine learning models can be trained on real data from the detector. I present the application of a Denoising Diffusion Model for stochastic event generation. Due to these models' ability to learn and improve, it is shown that they can more accurately describe the response of the detector than standard MC simulations do. The increase in quality of these new event generators allow for more accurate extraction of physical observables from experimental data, improving the precision of the results and better quantification of their uncertainties.