Speaker
Description
The quest towards probing the Higgs field in the High luminosity Large Hadron Collider (LHC) comes with many great challenges. In particular, the need to speed up the particle-detector simulations poses a roadblock, as projections show millions of CPU-years required to create simulated datasets. To tackle the problem of simulating particle-calorimeter interactions in the ATLAS detector at LHC we have developed a quantum-assisted deep generative model by combining quantum simulations with deep learning. In particular, we utilize D-wave’s Zephyr quantum annealer (QA) topology as a latent space prior using a variational autoencoder. We propose a robust method to generate conditioned samples using the quantum annealer leveraging flux bias to effectively increase the magnitude of the self-fields-to-interaction-energy ratio in the QA. We further propose a new, fast and robust method to estimate the effective inverse temperature in QAs. To benchmark our framework we use the CaloChallenge dataset, which has served as a catalyst for generative AI in high-energy calorimetry simulations. We compare our framework with 17 other frameworks which use generative AI and show that our framework is among the fastest and best in quality. The speed-up is three to six orders of magnitude compared to first-principles simulations used at CERN. We further assess the computational efficiency of our model in comparison to state-of-the-art generative models and first-principles approaches, demonstrating its potential for significantly accelerating high-energy physics simulations.
| Keyword-1 | quantum annealers |
|---|---|
| Keyword-2 | high-energy physi |
| Keyword-3 | generative AI |
