Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 120: Data, AI, Computing 9 (generative models & simulation)
T 120.2: Vortrag
Freitag, 8. März 2024, 09:15–09:30, Geb. 30.34: LTI
Accelerating event generation in Sherpa with deep learning with matrix element weight surrogates — •Tim Herrmann1, Timo Janßen2, Steffen Schumann2, and Frank Siegert1 — 1Technische Universität Dresden, Germany — 2Universität Göttingen, Germany
To calculate theory predictions for high energy physics (HEP) experiments, Monte Carlo (MC) methods are needed. Accelerating MC generation is needed to fulfil the future needs of HEP experiments. Unit-weight events are generated on particle level to make the most use of the following computationally expensive detector simulation. But making unit-weight events can also be time-consuming.
One way to save computation time during unit-weight event generation is to use a fast full matrix element weight surrogate. The surrogate is estimated via a deep neural network, which can be calculated much faster than the full matrix element. The estimate is corrected in a second step to get an unbiased prediction. This approach has already been shown to be effective. This work focuses on further optimizing it and working towards an implementation in an official Sherpa release.
Keywords: Monte Carlo; Sherpa; Deep learning; Matrix element weight surrogate; Machine learning