Abstract:
Generative machine learning models have been successfully used in order to speed up or augment many simulation tasks in particle physics, ranging from event generation to fast calorimeter simulation to many more. But even beyond simulation, generative approaches are seeing application for tasks like unfolding, anomaly detection, and even uncertainty quantification This indicates that generative models have great potential to become a mainstay in particle physics as a whole. One question that still needs to be addressed, however, is whether the data produced by a generative model can offer increased precision compared to the data the model was originally trained on. In other words, can one meaningfully draw more samples from a generative model than the ones it was trained with? We explore this using simplified models and demonstrate that generative models indeed have the capability to amplify data sets.
https://arxiv.org/abs/2008.06545
ZOOM LINK: