Research map / Learning theory and algorithms / Fairness
Fair Bayesian Data Selection via Generalized Discrepancy Measures
Yixuan Zhang, Jiabin Luo, Zhenggang Wang, Feng Zhou, Quyu Kong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add group-posterior alignment to a central distribution (Wasserstein) | BADS · UTKFace (bias 0.4) | accuracy: 0.751 -> 0.787 | Improved |
| add group-posterior alignment to a central distribution (Wasserstein) | BADS · LFW-A (bias 0.4) | DP: 0.217 -> 0.162 | Improved |
| add group-posterior alignment to a central distribution (Wasserstein) | BADS · LFW-A (bias 0.4) | accuracy: 0.834 -> 0.859 | Improved |
| add group-posterior alignment (Wasserstein) | BADS · FairFace (bias 0.4) | accuracy: 0.632 -> 0.662 | Improved |
| add group-posterior alignment (f-divergence) | BADS · UTKFace (bias 0.4) | EO: 0.303 -> 0.286 | No clear effect |
| use Fair-BADS-F instead of ERM | ERM · UTKFace (bias 0.2) | EO: 0.088 -> 0.286 | Hurt |
| replace explicit meta set with CLIP-RN50 zero-shot surrogate (ablation) | Fair-BADS-W · LFW-A (bias 0.4) | accuracy: 0.859 -> 0.847 | Hurt |
| replace explicit meta set with CLIP-RN50 zero-shot surrogate (ablation) | Fair-BADS · LFW-A | accuracy and fairness vs baselines | Improved |
| use ViT-B/16 backbone instead of ResNet-18 (ablation) | Fair-BADS-W · backbone ablation (bias 0.4) | accuracy: 0.820 -> 0.867 | Improved |
| use Wasserstein or MMD instead of f-divergence (ablation) | Fair-BADS · backbone ablation (bias 0.4) | fairness stability | Mixed |
| use fairness-unaware data selection (BLO, BADS) versus Fair-BADS | BLO, BADS · UTKFace, LFW-A, FairFace | fairness | Hurt |
| use fairness-constrained methods (FairBatch, FERM) | ERM · UTKFace (bias 0.4) | accuracy: 0.770 -> 0.721 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FairBatch — FairBatch: Batch Selection for Model Fairness (ICLR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Fairness-aware Bayes Optimal Functional Classification (NeurIPS 2025 · similarity 0.86)
- Achievable Fairness on Your Data With Utility Guarantees (NeurIPS 2024 · similarity 0.86)
- Navigating Towards Fairness with Data Selection (AAAI 2025 · similarity 0.86)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.86)
- Who’s the (Multi-)Fairest of Them All: Rethinking Interpolation-Based Data Augmentation Through the Lens of Multicalibration (AAAI 2025 · similarity 0.86)
- Fair Risk Control: A Generalized Framework for Calibrating Multi-group Fairness Risks (ICML 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
