Research map / Learning theory and algorithms / Fairness
Subgroup Robustness Grows On Trees: An Empirical Baseline Investigation
Joshua P Gardner, Zoran Popovi, Ludwig Schmidt
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 tree-based models (GBM, LightGBM, XGBoost, Random Forest) instead of robust or fairness-enhancing models | tabular subgroup robustness · eight tabular datasets | overall, worst-group accuracy and accuracy disparity | Improved |
| use XGBoost instead of Group DRO and chi^2 DRO | XGBoost · ACS Income, ACS Public Coverage, LARC | accuracy and worst-group accuracy | Improved |
| use DORO CVaR instead of tree models | robust models · Adult | maximum worst-group accuracy | No clear effect |
| use Inprocessing instead of tree models | fairness-enhancing models · Communities and Crime | maximum worst-group accuracy | No clear effect |
| select model by best CVaR instead of best accuracy (ablation) | DRO chi^2 and Group DRO · ACS Income and other datasets | worst-group accuracy | Hurt |
| select model by best CVaR instead of best accuracy (ablation) | tree-based models · ACS Income and other datasets | worst-group accuracy | No clear effect |
| use XGBoost instead of DRO chi^2 for hyperparameter sweep cost | tree models · ACS Income | sweep compute cost: 58 -> 1 | Improved |
| use tree-based models instead of others for hyperparameter sensitivity | tree models · all datasets | worst-group accuracy decline over ranked configurations | Improved |
| pair complementary metrics for model selection | all models · all datasets | metric correlation: 0.87, 0.99, 0.79 | Improved |
| select by robust risk (CVaR DORO), evaluate by accuracy | robust models · all datasets | accuracy | Hurt |
Closest papers by meaning
- Subgroups Matter for Robust Bias Mitigation (ICML 2025 · similarity 0.89)
- Multigroup Robustness (ICML 2024 · similarity 0.86)
- The Group Robustness is in the Details: Revisiting Finetuning under Spurious Correlations (NeurIPS 2024 · similarity 0.85)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.85)
- Enhancing Group Fairness in Online Settings Using Oblique Decision Forests (ICLR 2024 · similarity 0.85)
- Improving Subgroup Robustness via Data Selection (NeurIPS 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.
