Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative Diffusion Models for Fast Simulations of Particle Collisions at CERN

Mikołaj Kita, Jan Dubiński, Przemysław Rokita, Kamil Deja

TL;DR

The paper addresses the high computational cost of Geant4-based LHC detector simulations by introducing diffusion-model-based simulations for the ALICE Zero Degree Calorimeter. It demonstrates that conditional diffusion models achieve higher fidelity than existing VAEs/GANs and analyzes the trade-off between generation time and accuracy, highlighting Latent Diffusion Models for their rapid generation. The approach leverages a UNet backbone with particle-conditioned embeddings and explores DDIM sampling and latent-space diffusion to accelerate generation. The results indicate strong potential for diffusion-based fast simulations in high-energy physics, with Latent Diffusion Models offering notable speed advantages pending further fidelity improvements.

Abstract

In High Energy Physics simulations play a crucial role in unraveling the complexities of particle collision experiments within CERN's Large Hadron Collider. Machine learning simulation methods have garnered attention as promising alternatives to traditional approaches. While existing methods mainly employ Variational Autoencoders (VAEs) or Generative Adversarial Networks (GANs), recent advancements highlight the efficacy of diffusion models as state-of-the-art generative machine learning methods. We present the first simulation for Zero Degree Calorimeter (ZDC) at the ALICE experiment based on diffusion models, achieving the highest fidelity compared to existing baselines. We perform an analysis of trade-offs between generation times and the simulation quality. The results indicate a significant potential of latent diffusion model due to its rapid generation time.

Generative Diffusion Models for Fast Simulations of Particle Collisions at CERN

TL;DR

The paper addresses the high computational cost of Geant4-based LHC detector simulations by introducing diffusion-model-based simulations for the ALICE Zero Degree Calorimeter. It demonstrates that conditional diffusion models achieve higher fidelity than existing VAEs/GANs and analyzes the trade-off between generation time and accuracy, highlighting Latent Diffusion Models for their rapid generation. The approach leverages a UNet backbone with particle-conditioned embeddings and explores DDIM sampling and latent-space diffusion to accelerate generation. The results indicate strong potential for diffusion-based fast simulations in high-energy physics, with Latent Diffusion Models offering notable speed advantages pending further fidelity improvements.

Abstract

In High Energy Physics simulations play a crucial role in unraveling the complexities of particle collision experiments within CERN's Large Hadron Collider. Machine learning simulation methods have garnered attention as promising alternatives to traditional approaches. While existing methods mainly employ Variational Autoencoders (VAEs) or Generative Adversarial Networks (GANs), recent advancements highlight the efficacy of diffusion models as state-of-the-art generative machine learning methods. We present the first simulation for Zero Degree Calorimeter (ZDC) at the ALICE experiment based on diffusion models, achieving the highest fidelity compared to existing baselines. We perform an analysis of trade-offs between generation times and the simulation quality. The results indicate a significant potential of latent diffusion model due to its rapid generation time.
Paper Structure (9 sections, 2 equations, 5 figures, 2 tables)

This paper contains 9 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Zero Degree Calorimeter simulation
  4. Method
  5. Experiments and Results
  6. Evaluation
  7. Results
  8. Additional Experiments
  9. Conclusions

Figures (5)

  • Figure 1: ZDC cross-section with visible optic fibres grid (left) represented as a one-channel image (right)
  • Figure 2: Graphical presentation of forward and reverse diffusion pass. Image from DDPM2020
  • Figure 3: Accelerated generation with DDIM sampling method. Image from DDIM2020
  • Figure 4: Conditional Diffusion Model generations from the DDIM sampler at different number of inference steps
  • Figure 5: Conditional Diffusion Model generations from the DDPM sampler at different number of inference steps