Research map / Learning theory and algorithms / Fairness
Auditing Fairness by Betting
Ben Chugg, Santiago Cortes-Gomez, Bryan Wilder, Aaditya Ramdas
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 |
|---|---|---|---|
| Use sequential betting test instead of fixed-time permutation test with batches and no level correction (M1) | Betting audit (Algorithm 1) · Credit default, US census, insurance | false positive rate: > 0.5 | Improved |
| Use sequential betting test instead of Bonferroni-corrected batch permutation test (M2) | Betting audit (Algorithm 1) · Credit default and US census | stopping time under H1 at fixed FPR | Improved |
| Increase batch size k of permutation tests | Permutation test baselines · Credit default | stopping time: k = 1,500 gives tau = 1,500 minimum | Hurt |
| Switch from fair model to unfair random forest at time 400 (distribution shift) | Betting audit · Census data | type-I error control and rejection time | No clear effect |
| Apply batch permutation test M2 under distribution shift | Permutation test M2 · Census data with shift | false positive rate: FPR higher than 0.1 | Hurt |
| Collect audit data with randomized region-based policies pi1, pi2, pi3 instead of uniform sampling | Betting audit with weighted payoff (7) · Insurance data | rejection time | Hurt |
| Use betting with randomized policies instead of permutation tests on uniform data | Betting audit under policies pi1-pi3 · Insurance data | rejection time | Improved |
| Increase difference in means Delta between groups | Betting wealth process · Simulated Bernoulli streams | wealth growth and rejection speed | Improved |
| Add randomized final check K_tau >= U/alpha when budget runs out | Betting audit (Algorithm 1) · Theory | power | Improved |
Closest papers by meaning
- Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness (ICML 2018 · similarity 0.88)
- A Comprehensive View of Fairness through Distributional Stability (NeurIPS 2026 · similarity 0.87)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.86)
- Fair Off-Policy Learning from Observational Data (ICML 2024 · similarity 0.86)
- Cost Efficient Fairness Audit Under Partial Feedback (NeurIPS 2026 · similarity 0.86)
- Size-adaptive Hypothesis Testing for Fairness (NeurIPS 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.
