FairDiffuseVQVAE: Sampling-Time Fairness in Tabular Diffusion via Conditional Refinement of Vector-Quantized Latents

2026-07-30

AI Summary

Existing tabular data generation methods struggle to simultaneously achieve high fidelity and fairness. This work proposes a two-stage generative framework: in the first stage, a vector-quantized variational autoencoder (VQVAE) equipped with a row-level discriminator reconstructs high-fidelity data; in the second stage, a conditional diffusion model enforces demographic parity by uniformly sampling protected attributes during inference via classifier-free guidance, thereby embedding fairness into the generation mechanism without requiring explicit fairness constraints during training. Evaluated on Adult, Bank, and COMPAS datasets, the method substantially improves fairness—achieving a Demographic Parity Ratio of 0.702 (+47%) and an Equalized Odds Ratio of 0.686 (+100%)—while maintaining the lowest mean pairwise correlation error (0.034), at the cost of only approximately 15 AUC points.

Abstract

Synthetic tabular data is increasingly used in privacy-preserving data sharing, data augmentation, and to mitigate downstream classifier bias. State-of-the-art tabular diffusion models such as TabDDPM and TabSyn achieve excellent distributional fidelity but offer no mechanism for fairness; conversely, fairness-aware tabular generators (DECAF, FairTGAN, FairTabDDPM) impose explicit fairness penalties at training time, yielding modest fairness gains at substantial cost to either sample quality or downstream utility. We introduce FairDiffuseVQVAE, a two-stage architecture that decouples fidelity from fairness: a vector-quantized autoencoder with a row-level discriminator (Stage~1, no fairness terms) is followed by a DiffuseVAE-style continuous diffusion refiner that conditions on both the Stage-1 reconstruction and the protected attribute via classifier-free guidance (Stage~2). Fairness emerges as a property of the sampling distribution -- uniform sampling of the protected attribute at inference time enforces demographic parity by construction, rather than from competing loss terms. On the Adult, Bank and COMPAS datasets, FairDiffuseVQVAE achieves the highest mean Demographic Parity Ratio ($0.702$, $+47\%$ over FairTabDDPM) and Equalized Odds Ratio ($0.686$, $+100\%$). It also attains the lowest mean pair-wise correlation error ($0.034$) of any published method, while explicitly trading $\sim$$15$ AUC points for these fairness gains.

Problem

• tabular data generation
• fairness
• synthetic data
• demographic parity
• diffusion models

Innovation

• tabular diffusion
• fairness by sampling
• vector-quantized VAE
• classifier-free guidance
• demographic parity