Mutual information maximizing quantum generative adversarial networks.

Area of Science:

• Quantum Computing
• Artificial Intelligence
• Machine Learning

Background:

• Quantum generative adversarial networks (QGANs) show promise for quantum advantage in Noisy Intermediate-Scale Quantum (NISQ) computing.
• Existing QGANs face challenges like mode collapse and lack of explicit control over generated features.

Purpose of the Study:

• To introduce InfoQGAN, a novel quantum-classical hybrid model addressing QGAN limitations.
• To enhance feature control and mitigate mode collapse in quantum generative models.

Main Methods:

• Integration of InfoGAN principles into a QGAN architecture.
• Utilizing a variational quantum circuit for data generation.
• Employing a classical discriminator and a Mutual Information Neural Estimator (MINE) for optimizing latent code-sample mutual information.

Main Results:

• InfoQGAN effectively mitigates mode collapse in quantum generative models.
• Demonstrated robust feature disentanglement in the quantum generator.
• Showcased improved training stability and data augmentation performance via controlled feature generation.

Conclusions:

• InfoQGAN represents a significant advancement in quantum generative modeling for the NISQ era.
• The model enhances QGAN capabilities by enabling explicit control over generated data features.
• InfoQGAN provides a foundational approach for developing more sophisticated quantum machine learning applications.