Speaker
Lorenz Vogel
(Institute for Theoretical Physics, Heidelberg University)
Description
Calibrated learned uncertainties are a key requirement also for generative neural networks in LHC physics. For a toy model with an explicit likelihood we show how a heteroscedastic and a Bayesian normalizing flow learn the systematic and statistical uncertainties on the underlying phase space density. Without an explicit likelihood we train the heteroscedastic loss on a classifier-reweighted approximate generative network. We illustrate our comprehensive approach for top pair events and show how a conditional heteroscedastic flow propagates calibrated uncertainties to all phase space directions.
Authors
Anja Butter
(Centre National de la Recherche Scientifique (FR))
Lorenz Vogel
(Institute for Theoretical Physics, Heidelberg University)
Sascha Cassandra Diefenbacher
(Heidelberg University (DE))
Tilman Plehn