Hyperparameter tuning of BoloGAN: a CWGAN-GP model for fast simulation of the calorimeter of the ATLAS experiment

The analysis of current high energy experiments at the LHC requires simultaneous comparisons with simulations for the purposes of event reconstruction, background estimation and detector design and optimization, etc. Geant4 is the traditional simulation toolkit and has been widely used to carry these simulations. It has problems related to computational expense and slow speed. For fast simulation, Generative Models are being employed to approximate detector responses and event generation. BoloGAN is a fast simulation technique built using a conditional Wasserstein GAN. The aim of this thesis is to tune the hyperparameters of the model using random search based on the validation metric of reduced chi squares between Geant4 data and generated energy distributions by simulating the ATLAS calorimeter, on HPC systems. The tuning will be executed on LXBATCH at CERN remotely. This is done in view to optimize the performance of the model, to improve the detector response simulation, helping in the recostruction of the physics object in the events collected by the ATLAS experiment.