Research map / Learning theory and algorithms / Prediction and decision losses
Bitrate-Constrained DRO: Beyond Worst Case Robustness To Unknown Group Shifts
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 |
|---|---|---|---|
| Constrain the adversary with a bitrate penalty instead of using unconstrained CVaR DRO | Waterbirds · Waterbirds test | worst-group accuracy: 75.5 ± 0.4 -> 86.3 ± 0.3 | Helped |
| Constrain the adversary with a bitrate penalty instead of using unconstrained CVaR DRO | CelebA · CelebA test | worst-group accuracy: 64.7 ± 0.6 -> 80.9 ± 0.4 | Helped |
| Constrain the adversary with a bitrate penalty instead of using unconstrained CVaR DRO | CivilComments · CivilComments test | worst-group accuracy: 60.2 ± 0.8 -> 68.7 ± 0.9 | Helped |
| Use BR-DRO (L2) rather than unconstrained CVaR DRO | FMoW · FMoW test | worst-region accuracy: 31.0 ± 0.3 -> 32.3 ± 0.2 | Helped |
| Use BR-DRO (L2) rather than ERM | Camelyon17 · Camelyon17 test | average accuracy: 70.6 ± 1.6 -> 71.2 ± 1.0 | No clear effect |
| Use BR-DRO rather than JTT when training data has random label noise | Waterbirds and CelebA · Waterbirds and CelebA with random training-label noise | worst-group accuracy | Helped |
| Increase the VIB bitrate constraint strength (ablation) | BR-DRO minority identification · Waterbirds and CelebA | minority-group precision | Helped |
| Relax the bitrate constraint toward unconstrained CVaR DRO (ablation) | BR-DRO · Waterbirds and CelebA | minority-group precision | Hurt |
| Increase adversary capacity (weaken the capacity constraint) (ablation) | BR-DRO · Waterbirds and CelebA | worst-group accuracy | Helped |
| Use a sparse adversary rather than an unconstrained high-loss reweighting method (ablation) | Synthetic noisy-label data · Synthetic data | robust predictor recovery | Helped |
Closest papers by meaning
- MixMax: Distributional Robustness in Function Space via Optimal Data Mixtures (ICLR 2025 · similarity 0.85)
- Label-Efficient Group Robustness via Out-of-Distribution Concept Curation (CVPR 2024 · similarity 0.85)
- The Group Robustness is in the Details: Revisiting Finetuning under Spurious Correlations (NeurIPS 2024 · similarity 0.85)
- Rank-Constrained Adaptation for Reliable Real-World Performance (NeurIPS 2026 · similarity 0.85)
- Group Distributionally Robust Dataset Distillation with Risk Minimization (ICLR 2025 · similarity 0.84)
- Group-robust Sample Reweighting for Subpopulation Shifts via Influence Functions (ICLR 2025 · similarity 0.84)
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.