Where to Intervene? Benchmarking Fairness-Aware Learning on Differentially Private Synthetic Tabular Data

2026-07-08

AI Summary

This work addresses the tension between differential privacy (DP) and fairness in synthetic tabular data, where DP mechanisms may exacerbate disparities across groups and the efficacy of existing fairness interventions under DP constraints remains unclear. The study establishes the first comprehensive benchmark for fairness interventions in DP-synthesized data, systematically evaluating preprocessing, in-processing, and postprocessing strategies across multiple datasets, fairness metrics, and privacy budgets. Using the marginal DP synthesizer AIM, the authors compare four pipeline configurations on both original and synthetic data. Results demonstrate that applying DP alone degrades both utility and fairness, whereas incorporating fairness interventions—particularly postprocessing methods—partially restores fairness. Postprocessing consistently achieves robust fairness improvements across varying privacy budgets while preserving utility, highlighting its advantage in navigating the trade-offs among privacy, fairness, and utility.

Abstract

Machine learning models are increasingly deployed in high-stakes domains, raising concerns about both privacy and fairness. Differential Privacy (DP) has become a gold standard for privacy-preserving data analysis, while fairness-aware mechanisms aim to mitigate discrimination against underrepresented groups. However, these objectives can conflict: DP often amplifies disparities across demographic groups, and little is known about whether established fairness interventions remain effective under DP constraints. In this work, we present, to our knowledge, the first systematic evaluation of fairness interventions on differentially private synthetic tabular data. Our benchmark centers on the Adaptive Iterative Mechanism (AIM), identified as the state-of-the-art marginal-based DP synthesizer (Cormode et al. 2025). We thus evaluate fairness interventions across four datasets, multiple group fairness metrics, and three categories of mitigation strategies (pre-processing, in-processing, and post-processing) under a wide range of privacy budgets. We compare four pipeline configurations: (Baseline) training on original data; (DP-only) training on DP synthetic data; (Fair-only) applying fairness mechanisms on original data; and (DP+Fair) combining fairness mechanisms with DP synthetic data. Our results demonstrate that while DP alone can degrade both utility and fairness, applying fairness interventions can partially restore equitable outcomes. Among them, post-processing methods tend to provide more stable fairness-utility trade-offs across privacy budgets and synthesizers, achieving strong fairness improvements while preserving competitive utility relative to other intervention stages. We release all code, data, and experimental artifacts in an open-source repository to ensure full reproducibility and to support future research on the privacy-fairness-utility trade-off.

Problem

• differential privacy
• fairness-aware learning
• synthetic tabular data
• privacy-fairness trade-off
• fairness interventions

Innovation

• differential privacy
• fairness-aware learning
• synthetic tabular data
• post-processing fairness
• privacy-fairness trade-off