Heidelberg 2022 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 25: Data Analysis, Information Technology and Artificial Intelligence
T 25.3: Talk
Monday, March 21, 2022, 16:55–17:10, T-H38
Understanding Event-Generation Networks via Uncertainties — Marco Bellagente1, Manuel Haußmann2, •Michel Luchmann3, and Michel Luchmann4 — 1Institut für Theoretische Physik, Universität Heidelberg, Germany — 2Heidelberg Collaboratory for Image Processing, Universität Heidelberg, Germany — 3Institut für Theoretische Physik, Universität Heidelberg, Germany — 4Institut für Theoretische Physik, Universität Heidelberg, Germany
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 flows 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.