Хайдуров Руслан Александрович
Generative Models as an Approach for Fast Simulation of Showers Produced in Calorimeters for Experiments on Large Hadron Collider at CERN
Прикладная математика и информатика
In this paper, a new approach for calorimeter simulations is researched. Instead of using widespread numeric simulations frameworks, which are too slow to satisfy the growing demand on simulated events at CERN, this method uses deep generative models for faster simulations of electromagnetic showers. The methodology requires training and testing Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs). This approach has been validated on data simulated by the GEANT4 framework, which is used in CERN production, and show that the simulation procedure could be more than $10^4$ times faster without significant loss of quality. Trained models will also be tested on real simulation pipelines on the LHCb experiment.