A Survey of Quantum Generative Adversarial Networks: Architectures, Use Cases, and Real-World Implementations

TL;DR

This survey reviews the development, architectures, applications, and experimental progress of Quantum Generative Adversarial Networks, highlighting their potential for practical quantum advantage in various fields.

Contribution

It provides a comprehensive categorization of QGAN architectures, summarizes their real-world applications, and reviews experimental implementations on current quantum hardware.

Findings

QGAN architectures are categorized based on hybrid structures.

QGANs have been applied in image synthesis, medical data, and more.

Experimental demonstrations show progress but face hardware limitations.

Abstract

Quantum Generative Adversarial Networks (QGANs) have emerged as a promising direction in quantum machine learning, combining the strengths of quantum computing and adversarial training to enable efficient and expressive generative modeling. This survey provides a comprehensive overview of QGAN models, highlighting key advances from theoretical proposals to experimental realizations. We categorize existing QGAN architectures based on their quantum-classical hybrid structures and summarize their applications in fields such as image synthesis, medical data generation, channel prediction, software defect detection, and educational tools. Special attention is given to the integration of QGANs with domain-specific techniques, such as optimization heuristics, Wasserstein distance, variational circuits, and large language models. We also review experimental demonstrations on photonic and ion-trap quantum processors, assessing their feasibility under current hardware limitations. This survey aims to guide future research by outlining existing trends, challenges, and opportunities in developing QGANs for practical quantum advantage.