🔸 Title: FastCaloGAN : A fast calorimeter simulation tool exploiting Generative Adversarial Networks


🔸 Speaker: Dr Michele Faucci Giannelli (INFN Roma Tor Vergata, Italia)


🔸 Abstract: ATLAS relies on gargantuan amounts of simulated Monte Carlo events to carry out its wide physics programme. Currently, the detector simulation is performed predominantly using Geant4 but the CPU resources required are extraordinary. Therefore, the collaboration has being using and developing alternative fast simulation tools, focussing specifically on fast calorimeter shower simulation. FastCaloGAN is the most recent tool developed for this task and exploits Generative Adversarial Networks (GANs) for the generation of these showers. 300 GANs are used to parametrise the full calorimeter response; the challenges in training them and the performance of the stand-alone tool will be presented. These will motivate the choice to deploy FastCaloGAN as one of the tools used in AtlFast3 (AF3), the latest fast simulation application in ATLAS. AF3 meet the computing challenges and Monte Carlo needs for Run 3 and significantly improved the precision of the simulation that can now be used to simulate almost all physics processes.