Quantum Implementations of Generative Adversarial Networks (GANs) and a Comparative Analysis with the Classical Counterpart

Date
2023-02
Research Supervisor
Editor
Journal Title
Volume
Issue
Journal Title
Journal ISSN
Volume Title
Abstract
This thesis investigates the feasibility of using quantum computing to improve the performance of Generative Adversarial Networks (GANs). Two quantum GAN models, the Quantum Patch GAN and the QuGAN are compared to a classical GAN on the MNIST dataset, a widely used benchmark for machine learning algorithms. The classical GAN generated high-quality synthetic images with converging losses. The Quantum Patch GAN, however, showed a decline in performance after 700 epochs, but generated 0s from the set of 0s that the model was provided as input. The quantum state-based QuGAN performed best at the 50th epoch, generating decent-quality 9s and mildly recognizable 3s and 6s. The results of this research demonstrate the potential of quantum computing for generative modeling and offer new perspectives on the connection between quantum and classical computing in deep learning. Further exploration is needed to fully harness the advantages of quantum computing for GANs and advance the field of quantum computing itself. The potential impact of quantum GANs on real-world applications, such as drug discovery and materials science, cannot be overstated.
Description
Keywords
Citation
Department Name
Electrical and Computer Engineering
Publisher
North South University
Printed Thesis
DOI
ISSN
ISBN