Abstract:Current quantum generative adversarial networks (QGANs) still struggle with practical-sized data. First, many QGANs use principal component analysis (PCA) for dimension reduction, which, as our studies reveal, can diminish the QGAN's effectiveness. Second, methods that segment inputs into smaller patches processed by multiple generators face scalability issues. In this work, we propose LSTM-QGAN, a QGAN architecture that eliminates PCA preprocessing and integrates quantum long short-term memory (QLSTM) to ensure scalable performance. Our experiments show that LSTM-QGAN significantly enhances both performance and scalability over state-of-the-art QGAN models, with visual data improvements, reduced Frechet Inception Distance scores, and reductions of 5x in qubit counts, 5x in single-qubit gates, and 12x in two-qubit gates.

What problem does this paper attempt to address?

This paper attempts to solve several key problems faced by Quantum Generative Adversarial Networks (QGANs) when dealing with real - scale data: 1. **The influence of PCA pre - processing**: - Many existing QGANs use Principal Component Analysis (PCA) for dimension reduction. Research shows that PCA dominates the data processing process and weakens the effect of QGAN itself. Therefore, a method that does not rely on PCA is required to process high - dimensional data. 2. **The scalability problem of PatchGAN**: - PatchGAN achieves efficient processing by dividing the input into small pieces and having multiple generators process these pieces respectively. However, as the input size increases, PatchGAN requires more quantum resources, leading to serious scalability problems. For example, generating an MNIST image requires 56 sub - quantum generators and 280 qubits, which is infeasible in practical applications. In addition, as the number of qubits increases, the image quality generated by PatchGAN decreases significantly. To solve these problems, the paper proposes the LSTM - QGAN architecture with the following specific contributions: - **Preliminary analysis**: Through experiments on existing state - of - the - art QGANs (such as PCA - based QGANs and PatchGAN), the undisclosed limitations of PCA pre - processing and model scalability are revealed. - **Scalable architecture**: The LSTM - QGAN, a scalable QGAN architecture inspired by the recent Quantum Long - Short - Term Memory (QLSTM), is introduced. The LSTM - QGAN eliminates the need for PCA, maintains a constant NISQ resource usage as the input size grows, and can efficiently utilize the increasing quantum computing resources. - **Performance enhancement**: Evaluated on NISQ computers, the experimental results show that the LSTM - QGAN significantly enhances the generation performance and scalability. Compared with state - of - the - art QGANs, the LSTM - QGAN performs well in terms of visual data improvement, reduction in Fréchet Inception Distance (FID) scores, and reduction in the number of qubits and gates. ### Formula summary - **The loss function of the standard QGAN**: \[ \min_{\theta_g} \max_{\theta_d} L\{D_{\theta_d}(G_{\theta_g}(z)), D_{\theta_d}(x)\} \] - **The Wasserstein loss function**: \[ \min_{\theta_g} \max_{\theta_d} E_{x \sim P_r}[D_{\theta_d}(x)] - E_{\tilde{x} \sim P_g}[D_{\theta_d}(\tilde{x})] - \lambda \hat{L}_x \] where, \[ \hat{L}_x = E_{\hat{x} \sim \hat{P}_x} \left[ \left( \|\nabla_{\hat{x}} D_{\theta_d}(\hat{x})\|_2 - 1 \right)^2 \right] \] \(P_r\) and \(P_g\) represent the distributions of real data and generated data respectively, \(\hat{P}_x\) is the distribution sampled uniformly between \(P_r\) and \(P_g\), and \(\lambda\) is a constant. Through these improvements, the LSTM - QGAN not only improves the quality of the generated images but also greatly reduces the required quantum resources, thus providing a feasible solution for large - scale data processing in practical applications.

LSTM-QGAN: Scalable NISQ Generative Adversarial Network

Re-QGAN: an optimized adversarial quantum circuit learning framework

Variational Quantum Circuits Enhanced Generative Adversarial Network

Quantum generative adversarial networks with multiple superconducting qubits

Implementation Guidelines and Innovations in Quantum LSTM Networks

LatentQGAN: A Hybrid QGAN with Classical Convolutional Autoencoder

Survey of Quantum Generative Adversarial Networks (QGAN) to Generate Images

Experimental Quantum Generative Adversarial Networks for Image Generation

Quantum Generative Adversarial Networks in a Silicon Photonic Chip with Maximum Expressibility

Latent Style-based Quantum GAN for high-quality Image Generation

Entangling Quantum Generative Adversarial Networks

Quantum Wasserstein Generative Adversarial Networks.

QGAN: Quantized Generative Adversarial Networks.

A Survey of Recent Advances in Quantum Generative Adversarial Networks

Quantum Kernel-Based Long Short-term Memory

Dual-Parameterized Quantum Circuit GAN Model in High Energy Physics

A Hybrid Quantum-Classical Generative Adversarial Network for Near-Term Quantum Processors

Quantum Deep Generative Prior with Programmable Quantum Circuits

Dissipation-driven quantum generative adversarial networks

OrganiQ: Mitigating Classical Resource Bottlenecks of Quantum Generative Adversarial Networks on NISQ-Era Machines