TabularQGAN: Quantum Generative Model for Tabular Data Synthesis
An arXiv paper (2505.22533) presents TabularQGAN, an innovative quantum generative model designed for the synthesis of tabular data. This synthetic data can either complement or substitute datasets in situations where real-world data is limited or confidential. Given that enterprise data is mainly tabular and diverse—encompassing both categorical and numerical attributes—this development is relevant across sectors like healthcare, finance, and software. Unlike current quantum generative models that only address homogeneous data, TabularQGAN addresses this limitation through a quantum generative adversarial network architecture that incorporates adaptable data encoding and a novel quantum circuit ansatz. The model was evaluated using the MIMIC-III healthcare and Adult Census datasets, comparing its performance to top classical models such as CTGAN, CopulaGAN, VAE-GMM, and an LLM-based method via the be-GReaT framework.
Key facts
- TabularQGAN is a new quantum generative model for tabular data synthesis.
- The paper is available on arXiv under identifier 2505.22533.
- Synthetic data is useful when real-world data is scarce or private.
- Enterprise data is predominantly tabular and heterogeneous, containing categorical and numerical features.
- Existing quantum generative models are designed for homogeneous data.
- TabularQGAN uses a quantum GAN architecture with flexible data encoding and a novel quantum circuit ansatz.
- The model was tested on MIMIC-III healthcare and Adult Census datasets.
- Benchmarking includes CTGAN, CopulaGAN, VAE-GMM, and a be-GReaT-based LLM approach.
Entities
Institutions
- arXiv