Soft Margin Spectral Normalization for GANs

In this paper, we explore the use of Generative Adversarial Networks (GANs) to speed up the simulation process while ensuring that the generated results are consistent in terms of physics metrics. Our main focus is the application of spectral normalization for GANs to generate electromagnetic calorimeter (ECAL) response data, which is a crucial component of the LHCb. We propose an approach that allows to balance between model's capacity and stability during training procedure, compare it with previously published ones and study the relationship between proposed method's hyperparameters and quality of generated objects. We show that the tuning of normalization method's hyperparameters boosts the quality of generative model.