Aachen 2019 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 4: Deep Learning I

T 4.9: Talk

Monday, March 25, 2019, 18:00–18:15, H06

Parton showers with Generative Adversarial Networks — •Christof Sauer — Physikalisches Institut, Heidelberg, Deutschland

The prediction of physical processes are usually based on simulations – one example being parton showers. At present, the simulation of hadronic final states is done by dedicated software, such as Pythia, Herwig. This presentation intends to demonstrate a potential application of Generative Adversarial Networks (GANs) within the context of parton shower generation. Such machine learning techniques can be used to produce parton showers which are independent of any current shower model. It would allow to circumvent inherent problems in the simulation of parton showers. This method could be a applied in analyses that are too sensitive to parton shower effects in the modeling of the background and hence rely on an accurate background estimate.
As a first step, a network is trained on multijet events generated with Pythia, whereby the focus lies on training the network to produce realistic and consistent parton showers. The Monte Carlo samples serve as a surrogate to examine the applicability of this method under well controlled conditions before, subsequently, proceeding to use real data instead.