VAE-QWGAN: Improving Quantum GANs for High Resolution Image Generation

Aaron Mark Thomas,Sharu Theresa Jose

2024-09-16

Abstract:This paper presents a novel hybrid quantum generative model, the VAE-QWGAN, which combines the strengths of a classical Variational AutoEncoder (VAE) with a hybrid Quantum Wasserstein Generative Adversarial Network (QWGAN). The VAE-QWGAN integrates the VAE decoder and QGAN generator into a single quantum model with shared parameters, utilizing the VAE's encoder for latent vector sampling during training. To generate new data from the trained model at inference, input latent vectors are sampled from a Gaussian Mixture Model (GMM), learnt on the training latent vectors. This, in turn, enhances the diversity and quality of generated images. We evaluate the model's performance on MNIST/Fashion-MNIST datasets, and demonstrate improved quality and diversity of generated images compared to existing approaches.

Quantum Physics,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The paper attempts to address the problem of generating high-quality, high-resolution images under the limitations of current quantum technology, particularly NISQ devices. Specifically, the paper focuses on the challenges faced by Quantum Generative Adversarial Networks (QGANs) when dealing with high-dimensional classical datasets, such as generating low-quality samples and mode collapse issues.

To tackle these challenges, the authors propose a new hybrid quantum generative model called VAE-QWGAN. This model combines the advantages of classical Variational Autoencoders (VAE) and hybrid Quantum Wasserstein Generative Adversarial Networks (QWGAN) by integrating the VAE decoder and QGAN generator into a single quantum model through shared parameters and using the VAE encoder for latent vector sampling during training. Additionally, during the inference phase, the model uses a Gaussian Mixture Model (GMM) learned from the training latent vectors to generate new data, thereby improving the quality and diversity of the generated images.

The main contributions of the paper include:
1. **Model Design**: Proposing the VAE-QWGAN model, which combines the strengths of VAE and QWGAN.
2. **Training Method**: Ensuring stable training of the model by combining the loss functions of VAE and QWGAN.
3. **Performance Evaluation**: Conducting experiments on the MNIST and Fashion-MNIST datasets to validate the model's superiority in generating high-quality and diverse images.

In summary, the paper aims to address the quality and mode collapse issues of existing QGANs in generating high-resolution images by introducing the VAE-QWGAN model.