Research map / Learning theory and algorithms / Fairness
Post-processing for Individual Fairness
Felix Petersen, Debarghya Mukherjee, Yuekai Sun, Mikhail Yurochkin
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 |
|---|---|---|---|
| apply GLIF post-processing to BERT predictions | BERT on Bios · Bios | Pred. Consist.: 0.942 ± 0.002 -> 0.986 ± 0.002 | Improved |
| apply GLIF post-processing to BERT predictions | BERT on Bios · Bios | Test Acc.: 0.846 ± 0.003 -> 0.830 ± 0.004 | Hurt |
| replace the unnormalized Laplacian with the normalized random walk Laplacian (ablation) | GLIF on Bios · Bios | Pred. Consist.: 0.986 ± 0.002 -> 0.988 ± 0.002 | Improved |
| replace the unnormalized Laplacian with the normalized random walk Laplacian (ablation) | GLIF on Bios · Bios | Test Acc.: 0.830 ± 0.004 -> 0.834 ± 0.003 | Improved |
| apply GLIF post-processing to BERT predictions | BERT on Toxicity · Toxicity | Pred. Consist.: 0.614 ± 0.013 -> 0.835 ± 0.012 | Improved |
| apply GLIF post-processing to BERT predictions | BERT on Toxicity · Toxicity | Test Acc.: 0.809 ± 0.004 -> 0.803 ± 0.003 | Hurt |
| replace the unnormalized Laplacian with the normalized random walk Laplacian (ablation) | GLIF on Toxicity · Toxicity | Pred. Consist.: 0.835 ± 0.012 -> 0.844 ± 0.013 | Improved |
| replace the unnormalized Laplacian with the normalized random walk Laplacian (ablation) | GLIF on Toxicity · Toxicity | Test Acc.: 0.803 ± 0.003 -> 0.803 ± 0.003 | No clear effect |
| use GLIF rather than the global IF-constraints post-processing | GloVe sentiment model · Sentiment | accuracy at comparable fairness: 89.4% -> 95% | Improved |
| use GLIF post-processing | GloVe sentiment model · Sentiment | runtime: 422s -> 0.02s | Improved |
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.
- SenSEI — SenSeI: Sensitive Set Invariance for Enforcing Individual Fairness (ICLR 2021) · better on 3 of 4 shared results
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
- FRAPPÉ: A Group Fairness Framework for Post-Processing Everything (ICML 2024 · similarity 0.89)
- Task-Tailored Pre-Processing: Fair Downstream Supervised Learning (TPAMI 2026 · similarity 0.87)
- Average Individual Fairness: Algorithms, Generalization and Experiments (NeurIPS 2019 · similarity 0.87)
- Monotone Individual Fairness (ICML 2024 · similarity 0.86)
- Differentially Private Post-Processing for Fair Regression (ICML 2024 · similarity 0.85)
- Fairness Overfitting in Machine Learning: An Information-Theoretic Perspective (ICML 2025 · 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.
