|
Bayesian Generative Adversarial Networks via Gaussian Approximation for Tabular Data Synthesis
Bahrul Ilmi Nasution(a),(*), Mark Elliot(a), Richard Allmendinger(b)
Transactions on Data Privacy 19:3 (2026) 81 - 109
Abstract, PDF
(a) Department of Social Statistics, The University of Manchester, Oxford Rd, Manchester, M13 9PL, UK.
(b) Alliance Manchester Business School, The University of Manchester, Booth Street West, Manchester, M15 6PB, UK.
e-mail:firstname.lastname @manchester.ac.uk; firstname.lastname @manchester.ac.uk; firstname.lastname @manchester.ac.uk
|
|
Abstract
Generative Adversarial Networks (GAN) has been used in many studies to synthesise mixed tabular data. Conditional tabular GAN (CTGAN) has been the most popular variant but struggle to effectively navigate the risk-utility trade-off. Bayesian GAN has received less attention for tabular data, but have been explored with unstructured data such as images and text. The most used technique employed in Bayesian GAN is Markov Chain Monte Carlo (MCMC), but it is computationally intensive, particularly in terms of weight storage. In this paper, we introduce Gaussian Approximation of CTGAN (GACTGAN), an integration of the Bayesian posterior approximation technique using Stochastic Weight Averaging-Gaussian (SWAG) within the CTGAN generator to synthesise tabular data. Our results show that GACTGAN frequently improves tabular structure preservation and inferential utility, while maintaining comparable or lower disclosure risk in several datasets. These results highlight GACTGAN as a simpler, effective implementation of Bayesian tabular synthesis.
|
