Dortmund 2021 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 21: Data analysis, Information technology I

T 21.5: Vortrag

Montag, 15. März 2021, 17:00–17:15, Tu

Decoding γ-showers: Physics in the Latent Space of a BIB-AE Generative Network — •Erik Buhmann — Institut für Experimentalphysik, Universität Hamburg

With future collider experiments’ vast data collection capabilities and limited computing resources, interest in using generative neural networks for fast simulation of collider events is growing. In our previous study the Bounded Information Bottleneck Autoencoder (BIB-AE) showed state-of-the-art generation accuracy for photon showers in a high-granularity calorimeter, precisely modelling various global differential shower distributions. In this work we investigate how the BIB-AE encodes these physics information in the latent space for different model configurations. Our understanding of this latent space encoding allows us to propose methods to further optimize the generation performance of the BIB-AE model, namely specific hyperparameter optimization and an altered latent space sampling. In particular we were able to improve the modelling of the shower shape along the particle incident axis.