Research map / Probabilistic and scientific modeling / Causal discovery
Counterfactual Fairness with Imperfect Causal Graphs
Cong Su, Qiaoyu Tan, Carlotta Domeniconi, Lizhen Cui, Jun Wang, Guoxian Yu
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 |
|---|---|---|---|
| bound counterfactual quantities using local possible-descendant sets (CF-ICG) instead of using only definite non-descendants (Fair) | counterfactually fair classifier · Synthetic (20 nodes, 40 edges) | accuracy: 0.795 -> 0.835 | Improved |
| use CF-ICG instead of IFair for unidentifiable counterfactual quantities | counterfactually fair classifier · Synthetic (20 nodes, 40 edges) | unfairness: 0.136 -> 0.044 | Improved |
| apply CF-ICG fairness reconstruction on top of unconstrained baseline | 2-layer MLP · Adult | unfairness: 0.193 -> 0.043 | Improved |
| apply CF-ICG fairness reconstruction on top of unconstrained baseline | 2-layer MLP · Adult | accuracy: 0.787 -> 0.753 | Hurt |
| apply CF-ICG fairness reconstruction on top of unconstrained baseline | 2-layer MLP · Compas | unfairness: 0.232 -> 0.048 | Improved |
| apply CF-ICG fairness reconstruction on top of unconstrained baseline | 2-layer MLP · ACSPublicCoverage | unfairness: 0.247 -> 0.037 | Improved |
| use local structure of CF-ICG with different causal discovery algorithms (true CPDAG, PC, GES, MB-by-MB) (ablation) | CF-ICG · Synthetic (20 and 40 nodes) | accuracy and unfairness | No clear effect |
| increase CPDAG error rate from 0.2 to 0.6 (ablation) | CF-ICG vs IFair and Fair · Synthetic (20 nodes, 40 edges) | accuracy and unfairness | Mixed |
| use CF-ICG bounds on counterfactual fairness under unidentifiable contexts | CF-ICG · Synthetic (20 nodes, 40 edges) | bound coverage of ground truth: truth 0.186 within [0.173, 0.446] | Improved |
| exclude the sensitive attribute only (Unaware) rather than bounding counterfactual effects | 2-layer MLP · Synthetic (20 nodes, 40 edges) | unfairness: 0.170 vs 0.044 for CF-ICG | Hurt |
Closest papers by meaning
- Interventional Fairness on Partially Known Causal Graphs: A Constrained Optimization Approach (ICLR 2024 · similarity 0.91)
- A Local Method for Satisfying Interventional Fairness with Partially Known Causal Graphs (NeurIPS 2024 · similarity 0.90)
- Towards counterfactual fairness through auxiliary variables (ICLR 2025 · similarity 0.90)
- Counterfactual Fairness by Combining Factual and Counterfactual Predictions (NeurIPS 2024 · similarity 0.90)
- FairPFN: A Tabular Foundation Model for Causal Fairness (ICML 2025 · similarity 0.90)
- Fairness on Principal Stratum: A New Perspective on Counterfactual Fairness (ICML 2025 · similarity 0.88)
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.
