Research map / Learning theory and algorithms / Fairness
Providing Fair Recourse over Plausible Groups
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces group-level plausibility for algorithmic recourse, a plausibility-bias fairness metric, and Fair Feasible Training (FFT), a constrained training method that equalizes recourse plausibility across latent groups on Adult Income.
Problem
Algorithmic recourse methods return actions that ignore latent group characteristics not captured by the model, so recourse may be far less actionable or feasible for some groups (e.g., by sex or marital status).
Approach
Define group-level recourse plausibility as the coverage-weighted similarity of post-action features to clusters of approved profiles within a group, measure plausibility bias across groups, and train classifiers with a constraint that equalizes expected plausibility across groups (Fair Feasible Training).
Key results
- FFT reduced recourse redundancy by an average of 31.7% across recourse methods, at an average 24.6% increase in generation time.
- FFT produced a uni-modal plausibility distribution across groups, whereas standard training gave bimodal scores with female individuals facing higher feasibility costs.
- For both NN and LR models, FFT consistently gave recourse that was fairer in plausibility and overall cost than standard training.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train with Fair Feasible Training instead of standard training | Adult Income recourse models · Adult Income | Avg. Tim.: +24.6% increase in run time | Hurt |
| Train with Fair Feasible Training instead of standard training | Adult Income recourse models · Adult Income | Red.: an average 31.7% reduction | Helped |
| Train with Fair Feasible Training instead of standard training | Adult Income recourse models · Adult Income | Pro. | No clear effect |
| Train with Fair Feasible Training instead of standard training | Adult Income NN and LR models · Adult Income | plausibility fairness | Helped |
Surprising findings
- Overall proximity costs were not significantly affected by FFT, and proximity fairness improved even though it was not in the cost function.
Limitations
- Group-level plausibility may not ensure individual actionability.
- The approach may increase the cost of recourse.
- Evaluation limited to a single protected group split (sex) on one dataset.
Experimental setup
Datasets: Adult Income
Benchmarks: Adult Income
Metrics: success rate average time constraint violations redundancy proximity sparsity expected recourse cost expected plausibility group plausibility bias
Baselines: Wachter Growing Spheres Actionable Recourse FACE CCHVAE standard cross-entropy training
Builds on: Actionable Recourse CARLA recourse benchmark Gupta et al. 2019 fair recourse Karimi et al. plausibility
Closest papers by meaning
- Probabilistically Robust Recourse: Navigating the Trade-offs between Costs and Robustness in Algorithmic Recourse (ICLR 2023 · similarity 0.88)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.86)
- SafeAR: Safe Algorithmic Recourse by Risk-Aware Policies (AAAI 2024 · similarity 0.85)
- From Search to Sampling: Generative Models for Robust Algorithmic Recourse (ICLR 2025 · similarity 0.85)
- Prediction without Preclusion: Recourse Verification with Reachable Sets (ICLR 2024 · similarity 0.85)
- Fairness without Demographics through Shared Latent Space-Based Debiasing (AAAI 2024 · similarity 0.84)
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.