Karlsruhe 2024 – scientific programme

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

T: Fachverband Teilchenphysik

T 120: Data, AI, Computing 9 (generative models & simulation)

T 120.5: Talk

Friday, March 8, 2024, 10:00–10:15, Geb. 30.34: LTI

Study of data-augmentation for simulated ATLAS data sets using machine learning — •Lukas Vicenik, Boris Flach, and Andre Sopczak — Czech Technical University in Prague

Limitations on available simulated data sets have a strong impact on the uncertainty in searches for new particles and in precision measurements. This also applies to machine learning algorithms that aim to separate signal from background events. We propose to use Variational autoencoders (VAE) for augmenting and enriching simulated data sets. We consider two variants for training VAEs - the standard ELBO learning and a novel Nash equilibrium learning approach. The resulting generated data are validated and compared by studying their agreement with the original data sets.

Keywords: Data-augmentation; ATLAS; LHC; CERN