Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fast, accurate, and precise detector simulation with vision transformers

Luigi Favaro, Andrea Giammanco, Claudius Krause

TL;DR

This work tackles the need for fast yet accurate detector simulations by benchmarking vision-transformer–based generative models against Geant4. It decomposes shower generation into an energy component conditioned on the incident energy $E_{ ext{inc}}$ and a shape component conditioned on $E_{ ext{inc}}$ and energy ratios $u_i$, leveraging a patch-based 3D Vision Transformer to manage high-dimensional voxel data. The study compares discrete normalising flows (NFs) with coupling and rational-quadratic spline transforms to continuous normalising flows trained via Conditional Flow Matching (CFM), highlighting a speed–fidelity trade-off: NFs enable single-pass, very fast generation, while CFMs achieve higher fidelity at the cost of iterative sampling. Using CaloChallenge datasets, the authors show substantial speed advantages over Geant4 with meaningful fidelity, and provide public code for reproducibility and community-driven development of fast calorimeter simulations.

Abstract

The speed and fidelity of detector simulations in particle physics pose compelling questions about LHC analysis and future colliders. The sparse high-dimensional data, combined with the required precision, provide a challenging task for modern generative networks. We present a comparison between solutions with different trade-offs, including accurate Conditional Flow Matching and faster coupling-based Normalising Flows. Vision Transformers allows us to emulate the energy deposition from detailed Geant4 simulations. We evaluate the networks using high-level observables, neural network classifiers, and sampling timings, showing minimum deviations from Geant4 while achieving faster generation. We use the CaloChallenge benchmark datasets for reproducibility and further development.

Fast, accurate, and precise detector simulation with vision transformers

TL;DR

This work tackles the need for fast yet accurate detector simulations by benchmarking vision-transformer–based generative models against Geant4. It decomposes shower generation into an energy component conditioned on the incident energy and a shape component conditioned on and energy ratios , leveraging a patch-based 3D Vision Transformer to manage high-dimensional voxel data. The study compares discrete normalising flows (NFs) with coupling and rational-quadratic spline transforms to continuous normalising flows trained via Conditional Flow Matching (CFM), highlighting a speed–fidelity trade-off: NFs enable single-pass, very fast generation, while CFMs achieve higher fidelity at the cost of iterative sampling. Using CaloChallenge datasets, the authors show substantial speed advantages over Geant4 with meaningful fidelity, and provide public code for reproducibility and community-driven development of fast calorimeter simulations.

Abstract

The speed and fidelity of detector simulations in particle physics pose compelling questions about LHC analysis and future colliders. The sparse high-dimensional data, combined with the required precision, provide a challenging task for modern generative networks. We present a comparison between solutions with different trade-offs, including accurate Conditional Flow Matching and faster coupling-based Normalising Flows. Vision Transformers allows us to emulate the energy deposition from detailed Geant4 simulations. We evaluate the networks using high-level observables, neural network classifiers, and sampling timings, showing minimum deviations from Geant4 while achieving faster generation. We use the CaloChallenge benchmark datasets for reproducibility and further development.

Paper Structure

This paper contains 5 sections, 7 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. CaloChallenge datasets
  3. Generative Vision Transformers
  4. Results
  5. Conclusion

Figures (1)

  • Figure 1: Set of high-level features for the NF (top) and the CFM (bottom) for DS2. From left to right, we show the center of energy, the width of the center of energy, and the sparsity of calorimeter shower in a single detector layer. CFM plots are reproduced from Favaro:2024rle.