Flash Simulation at the LHC with Variational Autoencoders

We investigate the use of variational autoencoders in flash simulation of proton- proton collision events at the LHC. Given a dataset of over 2 million events, with 9 generator-level features and their software reconstructions, we train a variational autoencoder to reconstruct these features. Through analysis of dis- tributions and residuals, we assess the validity of this flash simulation method, identify weaknesses and propose solutions that could be applied in the future to a wide range of LHC experiments.