Understanding Event-Generation Networks via Uncertainties
This addresses the need for reliable uncertainty quantification in generative models for high-energy physics simulations, representing an incremental improvement in method application.
The paper tackled the problem of controlling generative neural networks and assigning uncertainties to their event output in LHC simulations, showing that Bayesian normalizing flow networks capture training uncertainties and convert them into event weight uncertainties.
Following the growing success of generative neural networks in LHC simulations, the crucial question is how to control the networks and assign uncertainties to their event output. We show how Bayesian normalizing flow or invertible networks capture uncertainties from the training and turn them into an uncertainty on the event weight. Fundamentally, the interplay between density and uncertainty estimates indicates that these networks learn functions in analogy to parameter fits rather than binned event counts.