Quantum Diffusion Model for Quark and Gluon Jet Generation

TL;DR

The paper tackles the computational bottlenecks of denoising diffusion models by introducing a fully quantum diffusion framework for high-energy physics jet generation, replacing Gaussian forward noise with Haar-uniform unitary scrambling and embedding a variational quantum circuit in the denoising network. It evaluates three variants—classical, hybrid, and fully quantum—on CMS quark–gluon jet data, using quantum embedding, forward diffusion, and a denoising quantum U-Net. Results show that quantum and hybrid configurations achieve competitive performance relative to a structurally similar classical baseline, suggesting potential practical advantages in quantum-accelerated generative modeling for physics data. The authors outline clear future directions, including expanding to all jet channels, testing alternative forward scramblings, and validating performance on quantum hardware to assess scalability and resilience to noise.

Abstract

Diffusion models have demonstrated remarkable success in image generation, but they are computationally intensive and time-consuming to train. In this paper, we introduce a novel diffusion model that benefits from quantum computing techniques in order to mitigate computational challenges and enhance generative performance within high energy physics data. The fully quantum diffusion model replaces Gaussian noise with random unitary matrices in the forward process and incorporates a variational quantum circuit within the U-Net in the denoising architecture. We run evaluations on the structurally complex quark and gluon jets dataset from the Large Hadron Collider. The results demonstrate that the fully quantum and hybrid models are competitive with a similar classical model for jet generation, highlighting the potential of using quantum techniques for machine learning problems.

Figures (6)

• Figure 1: The pipeline that the data goes through with all the possible classical-quantum combinations of the forward and backward diffusion process.
• Figure 2: A sample of an encoded jet image.
• Figure 3: Haar noise applied to one encoded sample of four channels.
• Figure 4: The losses and FID graphs of fully classical (a), hybrid (b), and fully quantum (c) models. For all models, MSE and Adam optimizer was used to reach convergence, and the FID function remained the same.
• Figure 5: Samples generated from random noise with the four encoded channels on the left four rows and the decoded image on the far right for each sample. Subfigure (a) uses the hybrid model and (b) uses the fully quantum model.