Generative Adversarial Networks for fast simulation

Federico Carminati,Gulrukh Khattak,Jean-Roch Vlimant,Konstantinos Samaras-Tsakiris,Vladimir Loncar,Maurizio Pierini,Thong Q Nguyen,Sofia Vallecorsa,Ricardo Brito Da Rocha

doi:10.1088/1742-6596/1525/1/012064

Abstract

Deep Learning techniques are being studied for different applications by the HEP community: in this talk, we discuss the case of detector simulation. The need for simulated events, expected in the future for LHC experiments and their High Luminosity upgrades, is increasing dramatically and requires new fast simulation solutions. Here we present updated results on the development of 3DGAN, one of the first examples using three-dimensional convolutional Generative Adversarial Networks to simulate high granularity electromagnetic calorimeters. In particular, we report on two main aspects: results on the simulation of a more general, realistic physics use case and on data parallel strategies to distribute the training process across multiple nodes on public cloud resources.

Highlights

Introduction and Related Work High EnergyPhysics (HEP) relies heavily on Monte Carlo simulation in order to model complex processes
Image generation is an important aspect of machine learning: a number of approaches exists including Generative Adversarial Networks (GAN)
GANs implement the idea of adversarial training for generating sharp and realistic images [7]; their application to High EnergyPhysics (HEP) simulation was introduced by the LAGAN [8] and the CaloGAN [9] models

Summary

Introduction

Introduction and Related Work High EnergyPhysics (HEP) relies heavily on Monte Carlo simulation in order to model complex processes. Detailed simulation of high granularity calorimeters is time-consuming, their output, a pattern of energy depositions in the different cells, can be interpreted as pixel intensities in a three-dimensional image.

Results

Conclusion