Reproduce “QTabGAN: A Hybrid Quantum-Classical GAN for Tabular Data Synthesis” on OpenResearch

About this paper

Subhangi Kumari, Rakesh Achutha, Vignesh Sivaraman

Researchers at the Indian Institute of Technology (BHU) and University of Cambridge developed QTabGAN, a hybrid quantum-classical GAN that synthesizes high-fidelity tabular data. This framework leverages variational quantum circuits within its generator to achieve lower machine learning utility differences and superior statistical similarity compared to state-of-the-art classical and existing quantum generative models.

Key insights

• Variational Quantum Circuits (VQCs) can effectively capture complex, high-dimensional probability distributions and inter-feature correlations inherent in tabular data by leveraging quantum superposition and entanglement.
• A hybrid quantum-classical architecture provides a practical balance, utilizing the expressive power of quantum circuits for distribution learning while relying on classical neural networks for scalable feature mapping and robust discrimination.
• The VQC's design, employing parameterized rotations and circular entanglement, optimizes quantum correlations and computational efficiency, making it more adaptable for near-term quantum hardware (NISQ devices) compared to fully quantum approaches.

Summary by alphaXiv.