Contributor
Chenguang Guan

Geant4-FastSim: Transformer-based architecture for fast shower simulation


Mentors
Dalila Salamani, Piyush Raikwar
Organization
CERN-HSF
Technologies
python, tensorflow, pytorch
Topics
machine learning, transformer, Generative Model, AI for Science
Calorimeter is one of the most important components of the Large Hadron Collider (LHC) experiments, detecting energy loss of particles after collision. However, current Geant4 simulation of showering process in the calorimeter is inherently slow for large amounts of events, especially after High Luminosity Upgrading in the future. Therefore, generative models and other machine learning techniques can be used to accelerate the Geant4 simulation. This project aims to improve the performances of the current VQ-VAE Transformer for particle showering and explore architectures beyond (VQ-)VAE and/or Transformer, including new position embedding, hierarchical attention, an-isotropic attention, Fourier Transform linear mixer, and etc. We expect to provide deliverables including notes of extended numeric experiment and integrating new models into the project code.