Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Variational Quantum Circuits Enhanced Generative Adversarial Network

Runqiu Shu, Xusheng Xu, Man-Hong Yung, Wei Cui

TL;DR

This work introduces QC-GAN, a hybrid quantum-classical generative framework that replaces part of the generator with a variational quantum circuit to improve image generation efficiency. Using a 5-qubit, depth-4 variational block and a classical discriminator, it demonstrates improved or comparable image quality (measured by FID) with far fewer trainable parameters and iterations on MNIST, compared to a classical GAN and to pathGAN at higher resolutions. The study highlights the resource-efficiency and expressivity gained from quantum circuits under NISQ constraints, while also noting training instability and sample diversity challenges inherent to quantum models. Overall, QC-GAN shows promise for quantum-assisted AI tasks, offering a concrete path to leveraging quantum resources for practical generative modeling with notable speedups and reduced parameter counts.

Abstract

Generative adversarial network (GAN) is one of the widely-adopted machine-learning frameworks for a wide range of applications such as generating high-quality images, video, and audio contents. However, training a GAN could become computationally expensive for large neural networks. In this work, we propose a hybrid quantum-classical architecture for improving GAN (denoted as QC-GAN). The performance was examed numerically by benchmarking with a classical GAN using MindSpore Quantum on the task of hand-written image generation. The generator of the QC-GAN consists of a quantum variational circuit together with a one-layer neural network, and the discriminator consists of a traditional neural network. Leveraging the entangling and expressive power of quantum circuits, our hybrid architecture achieved better performance (Frechet Inception Distance) than the classical GAN, with much fewer training parameters and number of iterations for convergence. We have also demonstrated the superiority of QC-GAN over an alternative quantum GAN, namely pathGAN, which could hardly generate 16$\times$16 or larger images. This work demonstrates the value of combining ideas from quantum computing with machine learning for both areas of Quantum-for-AI and AI-for-Quantum.

Variational Quantum Circuits Enhanced Generative Adversarial Network

TL;DR

This work introduces QC-GAN, a hybrid quantum-classical generative framework that replaces part of the generator with a variational quantum circuit to improve image generation efficiency. Using a 5-qubit, depth-4 variational block and a classical discriminator, it demonstrates improved or comparable image quality (measured by FID) with far fewer trainable parameters and iterations on MNIST, compared to a classical GAN and to pathGAN at higher resolutions. The study highlights the resource-efficiency and expressivity gained from quantum circuits under NISQ constraints, while also noting training instability and sample diversity challenges inherent to quantum models. Overall, QC-GAN shows promise for quantum-assisted AI tasks, offering a concrete path to leveraging quantum resources for practical generative modeling with notable speedups and reduced parameter counts.

Abstract

Generative adversarial network (GAN) is one of the widely-adopted machine-learning frameworks for a wide range of applications such as generating high-quality images, video, and audio contents. However, training a GAN could become computationally expensive for large neural networks. In this work, we propose a hybrid quantum-classical architecture for improving GAN (denoted as QC-GAN). The performance was examed numerically by benchmarking with a classical GAN using MindSpore Quantum on the task of hand-written image generation. The generator of the QC-GAN consists of a quantum variational circuit together with a one-layer neural network, and the discriminator consists of a traditional neural network. Leveraging the entangling and expressive power of quantum circuits, our hybrid architecture achieved better performance (Frechet Inception Distance) than the classical GAN, with much fewer training parameters and number of iterations for convergence. We have also demonstrated the superiority of QC-GAN over an alternative quantum GAN, namely pathGAN, which could hardly generate 1616 or larger images. This work demonstrates the value of combining ideas from quantum computing with machine learning for both areas of Quantum-for-AI and AI-for-Quantum.
Paper Structure (17 sections, 8 equations, 5 figures, 1 table)

This paper contains 17 sections, 8 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. PRELIMINARY
  3. Generative models
  4. Classical generative adversarial networks
  5. Variational quantum circuits
  6. QUANTUM GENERATIVE ADVERSARIAL NETWORK
  7. The architecture of the model
  8. Encoding
  9. Variational evolution
  10. Measurement and post-processing
  11. The training of the model
  12. NUMERICAL EXPERIMENTS
  13. Metics
  14. Experimental dataset and settings
  15. Experimental results and analysis
  16. ...and 2 more sections

Figures (5)

  • Figure 1: The algorithms based on PQCs are optimized in classical and quantum systems.
  • Figure 2: The overall architecture and training process of QC-GAN and the the classical GAN used in the experiment.
  • Figure 3: The FID scores of the classical GAN with 80, 768 parameters excluding the last layer and the QC-GAN with 65 quantum parameters. In addition, in the case of 768 parameters, the red line corresponds to an input size of 100 and the blue line corresponds to an input size of 5.
  • Figure 4: The comparison of patchGAN and QC-GAN for image generation, including the generation of handwritten digit 0 and 1 of size 8$\times$8 and 16$\times$16.
  • Figure 5: Handwritten digit generation of 28$\times$28 size using QC-GAN. (a)The noise input gradually changes to clear digits as the number of iterations increases. (b)The change of generator and discriminator loss curves during the training process.