Research map / Learning theory and algorithms / Prediction and decision losses
Focus on the Common Good: Group Distributional Robustness Follows
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 CGD instead of Group-DRO | classification with group annotations · Colored MNIST test | worst-group accuracy: 48.5 (11.5) -> 65.6 (5.9) | Helped |
| Use CGD instead of ERM-UW | classification with group annotations · WaterBirds test | worst-group accuracy: 86.9 (0.5) -> 88.9 (0.8) | Helped |
| Use CGD instead of Group-DRO | classification with group annotations · CelebA test | worst-group accuracy: 86.1 (0.9) -> 90.0 (0.8) | Helped |
| Use CGD instead of Group-DRO | classification with group annotations · MultiNLI test | worst-group accuracy: 76.6 (0.5) -> 76.1 (1.5) | No clear effect |
| Use CGD instead of ERM | classification with group annotations · Camelyon17 OOD | average accuracy: 70.3 (6.4) -> 69.4 (7.8) | No clear effect |
| Use CGD instead of Group-DRO | classification with group annotations · PovertyMap ID | worst-region Pearson r: 0.54 (0.11) -> 0.58 (0.05) | Helped |
| Use CGD instead of ERM | classification with group annotations · CivilComments ID | worst-group accuracy: 56.0 (3.6) -> 69.1 (1.9) | Helped |
| Focus updates on groups with high transfer rather than the highest-loss group | CGD · Noise-Simple | worst loss: 0.35 (0.03) -> 0.25 (0.02) | Helped |
| Focus updates on the central, high-transfer group | CGD · Rotation-Simple | worst loss: 0.77 (0.14) -> 0.59 (0.05) | Helped |
| Focus updates on the clean group with lower spurious correlation | CGD · Spurious-Simple | worst loss: 0.70 (0.16) -> 0.43 (0.06) | Helped |
| Use CGD instead of Group-DRO | classification with group annotations · Rotation-MNIST test | worst-group accuracy: 80.6 (1.1) -> 85.0 (0.7) | Helped |
Closest papers by meaning
- MixMax: Distributional Robustness in Function Space via Optimal Data Mixtures (ICLR 2025 · similarity 0.89)
- Efficient Algorithms for Empirical Group Distributionally Robust Optimization and Beyond (ICML 2024 · similarity 0.88)
- The Group Robustness is in the Details: Revisiting Finetuning under Spurious Correlations (NeurIPS 2024 · similarity 0.88)
- Rank-Constrained Adaptation for Reliable Real-World Performance (NeurIPS 2026 · similarity 0.88)
- Improving Subgroup Robustness via Data Selection (NeurIPS 2024 · similarity 0.87)
- Group Distributionally Robust Dataset Distillation with Risk Minimization (ICLR 2025 · similarity 0.87)
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.