Quantum Walks-Based Adaptive Distribution Generation with Efficient CUDA-Q Acceleration

Yen-Jui Chang, Wei-Ting Wang, Chen-Yu Liu, Yun-Yuan Wang, Ching-Ray ChangarXiv 2025 · DOI · arXiv

Abstract

State preparation remains a significant hurdle in quantum computing, particularly for algorithms requiring specific initial probability distributions. This paper introduces an Adaptive Distribution Generator (ADG) that leverages variational quantum circuits (VQCs) and split-step quantum walks (SSQWs) to "sculpt" high-precision distributions. By utilizing NVIDIA’s CUDA-Q framework for GPU-accelerated simulation, the researchers demonstrate the efficient generation of 1D financial models and 2D spatial patterns, providing a scalable path for quantum data loading.

Introduction

In quantum algorithms ranging from financial risk analysis to machine learning, the ability to load classical data as a quantum state is essential. However, traditional state preparation methods often struggle with precision or computational complexity as the target distribution grows in detail. The proposed Adaptive Distribution Generator (ADG) addresses this by using the natural spreading and interference properties of a quantum walker. By tuning the "coin" that dictates the walker's movement, the system can iteratively learn and reproduce complex target distributions.

Methodology: The Hybrid ADG Framework

The ADG framework is built upon a hybrid quantum-classical loop. At its core are Split-Step Quantum Walks (SSQWs), which provide finer control over state evolution than standard quantum walks by breaking movement into discrete sub-steps.

Hybrid quantum-classical framework for the Parameterized SSQW layer.

The "coin" operations within the circuit are implemented as Variational Quantum Circuits (VQCs) using tunable U3 gates. A classical optimizer adjusts the gate parameters (angles $\theta, \phi, \lambda$) to minimize the divergence between the generated quantum state's probability distribution and the target classical data. To handle the significant linear algebra required for these simulations, the team employed the CUDA-Q platform, offloading the optimization loop to NVIDIA GPUs for substantial speedups.

2D Pattern Generation and Entanglement

Moving beyond 1D lines to 2D grids requires capturing spatial correlations. The authors achieved this by entangling the coin registers of two separate quantum walkers. This entanglement allows the walkers to move in correlated ways across X and Y axes, effectively "drawing" 2D patterns.

Parameterized Entangled-SSQW layer for generating 2D spatial distributions.

As shown in the figure above, the entangled-SSQW layer maps the quantum state to a 2D intensity grid $I(x_i, y_i)$, where the probability amplitudes represent pixel values. This approach was successfully used to reproduce 8×8 pixel representations of MNIST-style digits, proving that entangled quantum walks can serve as a foundation for quantum image processing.

Results and Applications

The ADG was benchmarked across several domains:

  • Statistical Distributions: The model matched standard distributions (Beta, Binomial, Bimodal, Exponential, and Poisson) with high fidelity.
  • Finance: In tests involving NVIDIA (NVDA) stock return profiles and European call option pricing, the quantum-simulated prices closely tracked analytical Black-Scholes benchmarks, particularly for "at-the-money" strikes.
  • Scalability: The use of CUDA-Q and GPU acceleration allowed the researchers to handle the computational overhead of the iterative optimization process, which would be significantly slower on standard CPU-based simulators.

Conclusion

The QWs-based ADG offers a flexible and accurate method for quantum state preparation. While current simulations on 5-qubit systems face some discretization errors in sparse data regions, the integration of GPU acceleration demonstrates a viable path for scaling hybrid workflows. Future work will focus on multi-coin configurations and testing the framework on physical NISQ hardware with advanced error mitigation strategies.

Summary updated

Abstract

We present a novel Adaptive Distribution Generator that leverages a quantum walks-based approach to generate high precision and efficiency of target probability distributions. Our method integrates variational quantum circuits with discrete-time quantum walks, specifically, split-step quantum walks and their entangled extensions, to dynamically tune coin parameters and drive the evolution of quantum states towards desired distributions. This enables accurate one-dimensional probability modeling for applications such as financial simulation and structured two-dimensional pattern generation exemplified by digit representations(0~9). Implemented within the CUDA-Q framework, our approach exploits GPU acceleration to significantly reduce computational overhead and improve scalability relative to conventional methods. Extensive benchmarks demonstrate that our Quantum Walks-Based Adaptive Distribution Generator achieves high simulation fidelity and bridges the gap between theoretical quantum algorithms and practical high-performance computation.
BibTeX
@misc{chang2025quantum,
  author = {Yen-Jui Chang and Wei-Ting Wang and Chen-Yu Liu and Yun-Yuan Wang and Ching-Ray Chang},
  title = {{Quantum Walks-Based Adaptive Distribution Generation with Efficient CUDA-Q Acceleration}},
  year = {2025},
  eprint = {2504.13532},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2504.13532},
}
Figure