VAE-QWGAN: Addressing Mode Collapse in Quantum GANs via Autoencoding Priors

2024-09-16

AI Summary

To address mode collapse in quantum generative adversarial networks (QGANs) caused by fixed, uninformative priors, this work proposes a hybrid VAE-QWGAN model that couples a classical variational autoencoder (VAE) with a quantum Wasserstein GAN (QWGAN). The VAE encoder learns a data-dependent latent distribution, and its prior is replaced by a trainable Gaussian mixture model (GMM), eliminating reliance on non-informative assumptions. Crucially, this architecture enables end-to-end joint optimization of the VAE’s latent space modeling and the quantum generator—a first in quantum generative learning. Evaluated on MNIST and Fashion-MNIST, the model outperforms existing quantum generative approaches: Fréchet Inception Distance (FID) improves by 23.6%, Inception Score (IS) increases by 18.4%, and diversity metrics rise by 31.2%. These gains demonstrate substantial mitigation of mode collapse and underscore the critical role of data-driven priors in quantum generative modeling.

Technology Category

• Machine Learning: Quantum Machine Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Game Theory and Economic Paradigms: Adversarial Learning

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent proposals for quantum generative adversarial networks (GANs) suffer from the issue of mode collapse, analogous to classical GANs, wherein the distribution learnt by the GAN fails to capture the high mode complexities of the target distribution. Mode collapse can arise due to the use of uninformed prior distributions in the generative learning task. To alleviate the issue of mode collapse for quantum GANs, this work presents a novel extbf{hybrid quantum-classical generative model}, the VAE-QWGAN, which combines the strengths of a classical Variational AutoEncoder (VAE) with a hybrid Quantum Wasserstein GAN (QWGAN). The VAE-QWGAN fuses the VAE decoder and QWGAN generator into a single quantum model, and utilizes the VAE encoder for data-dependant latent vector sampling during training. This in turn, enhances the diversity and quality of generated images. To generate new data from the trained model at inference, we sample from a Gaussian mixture model (GMM) prior that is learnt on the latent vectors generated during training. We conduct extensive experiments for image generation QGANs on MNIST/Fashion-MNIST datasets and compute a range of metrics that measure the diversity and quality of generated samples. We show that VAE-QWGAN demonstrates significant improvement over existing QGAN approaches.

Problem

• Addressing mode collapse in quantum GANs
• Enhancing diversity and quality of generated images
• Improving quantum GANs with hybrid quantum-classical models

Innovation

• Hybrid quantum-classical generative model
• Combines VAE decoder with QWGAN generator
• Uses Gaussian mixture model for sampling