Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Applying generative neural networks for fast simulations of the ALICE (CERN) experiment

Maksymilian Wojnar

TL;DR

This work addresses the need for fast, high-fidelity simulations of the ALICE ZDC neutron detector by evaluating state-of-the-art generative neural networks. It methodically compares autoencoders, GANs, vector-quantized models, and diffusion frameworks across ViT/MLP-Mixer architectures, focusing on reconstruction quality, sample diversity, and generation speed, with a targeted 5 ms per-sample performance. The study finds diffusion models deliver the best Wasserstein fidelity (score around 3.15) but at slower generation times, while VQ-GAN offers a practical balance between quality and throughput; classical GAN with postprocessing also achieves strong results. The results inform concrete recommendations for fast ZDC simulations, including architecture choices, codebook management, sampling strategies, and future directions toward incorporating physical losses and latent-space optimizations, thereby enabling efficient, physics-consistent surrogate simulations for CERN-scale experiments.

Abstract

This thesis investigates the application of state-of-the-art advances in generative neural networks for fast simulation of the Zero Degree Calorimeter (ZDC) neutron detector in the ALICE experiment at CERN. Traditional simulation methods using the GEANT Monte Carlo toolkit, while accurate, are computationally demanding. With increasing computational needs at CERN, efficient simulation techniques are essential. The thesis provides a comprehensive literature review on the application of neural networks in computer vision, fast simulations using machine learning, and generative neural networks in high-energy physics. The theory of the analyzed models is also discussed, along with technical aspects and the challenges associated with a practical implementation. The experiments evaluate various neural network architectures, including convolutional neural networks, vision transformers, and MLP-Mixers, as well as generative frameworks such as autoencoders, generative adversarial networks, vector quantization models, and diffusion models. Key contributions include the implementation and evaluation of these models, a significant improvement in the Wasserstein metric compared to existing methods with a low generation time of 5 milliseconds per sample, and the formulation of a list of recommendations for developing models for fast ZDC simulation. Open-source code and detailed hyperparameter settings are provided for reproducibility. Additionally, the thesis outlines future research directions to further enhance simulation fidelity and efficiency.

Applying generative neural networks for fast simulations of the ALICE (CERN) experiment

TL;DR

This work addresses the need for fast, high-fidelity simulations of the ALICE ZDC neutron detector by evaluating state-of-the-art generative neural networks. It methodically compares autoencoders, GANs, vector-quantized models, and diffusion frameworks across ViT/MLP-Mixer architectures, focusing on reconstruction quality, sample diversity, and generation speed, with a targeted 5 ms per-sample performance. The study finds diffusion models deliver the best Wasserstein fidelity (score around 3.15) but at slower generation times, while VQ-GAN offers a practical balance between quality and throughput; classical GAN with postprocessing also achieves strong results. The results inform concrete recommendations for fast ZDC simulations, including architecture choices, codebook management, sampling strategies, and future directions toward incorporating physical losses and latent-space optimizations, thereby enabling efficient, physics-consistent surrogate simulations for CERN-scale experiments.

Abstract

This thesis investigates the application of state-of-the-art advances in generative neural networks for fast simulation of the Zero Degree Calorimeter (ZDC) neutron detector in the ALICE experiment at CERN. Traditional simulation methods using the GEANT Monte Carlo toolkit, while accurate, are computationally demanding. With increasing computational needs at CERN, efficient simulation techniques are essential. The thesis provides a comprehensive literature review on the application of neural networks in computer vision, fast simulations using machine learning, and generative neural networks in high-energy physics. The theory of the analyzed models is also discussed, along with technical aspects and the challenges associated with a practical implementation. The experiments evaluate various neural network architectures, including convolutional neural networks, vision transformers, and MLP-Mixers, as well as generative frameworks such as autoencoders, generative adversarial networks, vector quantization models, and diffusion models. Key contributions include the implementation and evaluation of these models, a significant improvement in the Wasserstein metric compared to existing methods with a low generation time of 5 milliseconds per sample, and the formulation of a list of recommendations for developing models for fast ZDC simulation. Open-source code and detailed hyperparameter settings are provided for reproducibility. Additionally, the thesis outlines future research directions to further enhance simulation fidelity and efficiency.
Paper Structure (60 sections, 37 equations, 35 figures, 24 tables)

This paper contains 60 sections, 37 equations, 35 figures, 24 tables.

Table of Contents

  1. Introduction
  2. Motivation and scope of work
  3. Contributions
  4. Thesis overview
  5. Related work
  6. Generative neural networks
  7. Autoencoders
  8. Generative adversarial networks
  9. Autoregressive models
  10. Normalizing flows
  11. Vector quantization
  12. Diffusion models
  13. Fast simulations
  14. Machine learning for fast physical simulations
  15. Generative neural networks for fast simulations in CERN
  16. ...and 45 more sections

Figures (35)

  • Figure 1: Scheme of the CERN complex in 2022 lhc_photo.
  • Figure 2: The ZDC neutron detector construction zdc_image_1zdc_image_2.
  • Figure 3: Proton tracks passing through the ALICE detector simulated using GEANT zdc_technical_report.
  • Figure 4: Example ZDC neutron detector responses simulated using Monte Carlo techniques.
  • Figure 5: The variational autoencoder design.
  • ...and 30 more figures