Validation of Deep Convolutional Generative Adversarial Networks for High Energy Physics Calorimeter Simulations

Abstract

In particle physics the simulation of particle transport through detectors requires an enormous amount of computational resources, utilizing more than 50% of the resources of the CERN Worldwide Large Hadron Collider Grid. This challenge has motivated the investigation of different, faster approaches for replacing the standard Monte Carlo simulations. Deep Learning Generative Adversarial Networks are among the most promising alternatives. Previous studies showed that they achieve the necessary level of accuracy while decreasing the simulation time by orders of magnitudes. In this paper we present a newly developed neural network architecture which reproduces a three-dimensional problem employing 2D convolutional layers and we compare its performance with an earlier architecture consisting of 3D convolutional layers. The performance evaluation relies on direct comparison to Monte Carlo simulations, in terms of different physics quantities usually employed to quantify the detector response. We prove that our new neural network architecture reaches a higher level of accuracy with respect to the 3D convolutional GAN while reducing the necessary computational resources. Calorimeters are among the most expensive detectors in terms of simulation time. Therefore we focus our study on an electromagnetic calorimeter prototype with a regular highly granular geometry, as an example of future calorimeters.

Hadrons, better, faster, stronger

“…Motivated by the large fraction of resources already consumed by calorimeter simulation [5], and the expected increase due to higher granularities and luminosities, the precise and fast simulation of calorimeters is a primary topic of research [2,[6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23] 4 .…”

New angles on fast calorimeter shower simulation

“…The second physical distribution relating to the shower development along the depth of the calorimeter that was investigated is the longitudinal profile. 7 The JSD values between BIB-AE and GEANT4 distributions for each combination of incident particle energy and angle are shown in table 3. At simulation level, the best (90 GeV, 85 degrees) and worst (20 GeV, 40 degrees) performing combinations for energy and angle are highlighted in bold, and the distributions shown in figure 16.…”

“…A promising alternative approach to potentially speed up simulation is to use a generative model based surrogate simulator. To this end, a plethora of different generative models have been proposed for the task, including Generative Adversarial Networks (GANs) [3][4][5][6][7][8][9][10], Bounded Information Bottleneck Autoencoders (BIB-AEs) [11,12], Wasserstein GANs (WGANS) [13,14], Normalising Flows [15][16][17] and Score-Based Models [18].…”

Evaluating Generative Adversarial Networks for particle hit generation in a cylindrical drift chamber using Fréchet Inception Distance

“…Evaluating the GAN is crucial to ensure that it can be used to augment the simulated data and enable accurate sensitivity estimates. In previous works, deep generative models were evaluated by measuring the similarity of real and generated distributions of explicitly or implicitly learned features [6][7][8], using the accuracy of a classifier as a proxy [6,9] or by measuring Fréchet Inception Distance [10] (FID) in a latent space [7,11]. The latter method has the advantage to directly compare real and fake samples, but necessitates a fully trained inception network, which itself requires a lot of resources.…”