Research map / Learning theory and algorithms / Prediction and decision losses
Certifying Robustness to Programmable Data Bias in Decision Trees
Anna P. Meyer, Aws Albarghouthi, Loris D'Antoni
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 |
|---|---|---|---|
| increase label-flipping bias from 0.05% to 1% of training set | Antidote-P · COMPAS | certified robustness (%): 81.9 -> 3.0 | Hurt |
| increase missing-data bias from 0.05% to 0.4% | Antidote-P · Adult Income | certified robustness (%): 96.0 -> 60.9 | Hurt |
| switch from missing-data to label-flipping bias model at 1.0% bias | Antidote-P · Drug Consumption | certified robustness (%): 85.1 -> 7.1 | Hurt |
| use label-flipping instead of missing data at 0.4% bias | Antidote-P · Adult Income | certified robustness (%): 60.9 -> 34.8 | Hurt |
| use targeted bias model instead of untargeted label-flipping at 0.4% | Antidote-P · COMPAS | certified robustness (%): 20.6 -> 50.5 | Improved |
| use targeted missing-data model instead of untargeted at 1.0% | Antidote-P · Adult Income | certified robustness (%): 31.6 | Improved |
| combine missing and fake bias (0.1% each) instead of 0.1% label-flipping | Antidote-P · COMPAS | certified robustness (%): 71.5 -> 50.5 | Hurt |
| compare dataset: Drug Consumption vs Adult Income vs COMPAS under label-flipping 0.2% | Antidote-P · three datasets | certified robustness (%): 94.5 vs 70.2 vs 47.8 | Mixed |
| White women vs Black people subgroup under label-flipping 0.4% | Antidote-P · COMPAS | certified robustness (%): 50.4 vs 0 | Mixed |
| White women vs White men under label-flipping 0.4% | Antidote-P · COMPAS | certified robustness (%): 50.4 vs 38.8 | Mixed |
| targeted bias on (Black, positive) at high bias levels, White vs Black test samples | Antidote-P · COMPAS | certified robustness (%): 32.9 vs 0 | Mixed |
| targeted predicate (White, negative) instead of (Black, positive) | Antidote-P · COMPAS | demographic robustness gap | Improved |
Closest papers by meaning
- Learning Optimal and Fair Decision Trees for Non-Discriminative Decision-Making (AAAI 2019 · similarity 0.88)
- An Information-theoretic Framework for Auditing Unfairness in Training Data (NeurIPS 2026 · similarity 0.87)
- Mitigating Label Bias in Machine Learning: Fairness through Confident Learning (AAAI 2024 · similarity 0.86)
- SEBRA : Debiasing through Self-Guided Bias Ranking (ICLR 2025 · similarity 0.85)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.85)
- Robust Loss Functions for Training Decision Trees with Noisy Labels (AAAI 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.
