Sparse autoregressive models for scalable generation of sparse images in particle physics

Generation of simulated data is essential for analysis in particle physics, but current Monte Carlo methods are very computationally expensive. Deep-learning-based generative models have successfully generated at lower cost, struggle when the sparse. We introduce a novel deep sparse autoregressive model (SARM) that explicitly learns sparseness with tractable likelihood, making it more stable and interpretable compared to adversarial networks (GANs) other methods. In two case studies, we compare SARM GAN nonsparse model. As quantitative measure performance, compute Wasserstein distance (${W}_{p}$) between distributions physical quantities calculated on images training images. first study, featuring jets which 90% pixels zero valued, produces ${W}_{p}$ scores 24%--52% better than obtained state-of-the-art models. second calorimeter vicinity muons where 98% 66%--68% better. Similar observations made metrics confirm usefulness physics.