A universal vision transformer for fast calorimeter simulations
Researchers have developed a universal vision transformer (ViT) capable of simulating calorimeter responses in high-energy physics detectors with high accuracy and speed. This new architecture, building on CaloDREAM, demonstrates robustness across various geometries and detectors, generating electromagnetic and hadronic showers with minimal deviation from established simulation methods like Geant4. The ViT achieves simulation times in the tens to hundreds of milliseconds on a single GPU, and pre-training followed by fine-tuning further enhances efficiency and fidelity. AI
IMPACT Enables faster and more accurate simulations for high-energy physics experiments, potentially accelerating discovery.