Research map / Label-efficient and robust learning / Test-time adaptation
Distributionally Robust Models with Parametric Likelihood Ratios
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 parametric likelihood-ratio adversary instead of P-DRO | DRO · BiasedSST | robust accuracy: 34.98 ± 9.39 -> 50.70 ± 7.33 | Helped |
| use parametric likelihood-ratio adversary instead of P-DRO | DRO · FDCL18 | robust accuracy: 30.25 ± 10.13 -> 53.52 ± 1.66 | Helped |
| use RP-DRO instead of the strongest listed nonparametric baseline | DRO · Waterbirds | robust accuracy: 72.21 ± 0.95 -> 73.49 ± 3.01 | Helped |
| use RP-DRO instead of the strongest listed nonparametric baseline | DRO · CelebA | robust accuracy: 43.33 ± 3.58 -> 55.78 ± 9.15 | Helped |
| use RP-DRO rather than ERM | classifier · BiasedSST | average accuracy: 95.09 ± 0.16 -> 86.60 ± 2.96 | Hurt |
| use RP-DRO rather than ERM | classifier · FDCL18 | average accuracy: 81.56 ± 0.26 -> 76.62 ± 1.43 | Hurt |
| use RP-DRO rather than NonParam-KL | classifier · Waterbirds | average accuracy: 90.54 ± 0.72 -> 90.15 ± 0.74 | No clear effect |
| use RP-DRO rather than ERM | classifier · CelebA | average accuracy: 95.89 ± 0.05 -> 93.10 ± 3.87 | Hurt |
| add label noise at p_noise=0.1 (ablation) | RP-DRO vs nonparametric DRO on BiasedSST · BiasedSST with label noise | average test accuracy: 85 -> 65; around 5-point loss for RP-DRO | Mixed |
| allow the adversary more gradient steps per classifier update (ablation) | RP-DRO toy setting · toy two-domain classification | robust accuracy | Hurt |
| use batch-level normalization instead of self-normalization (ablation) | RP-DRO · BiasedSST | robust accuracy | Helped |
| use batch-level normalization instead of self-normalization (ablation) | RP-DRO with oracle stopping · BiasedSST | robust accuracy | No clear effect |
Closest papers by meaning
- Learning Distributionally Robust Models at Scale via Composite Optimization (ICLR 2022 · similarity 0.91)
- DORO: Distributional and Outlier Robust Optimization (ICML 2021 · similarity 0.90)
- Large-Scale Non-convex Stochastic Constrained Distributionally Robust Optimization (AAAI 2024 · similarity 0.87)
- Multiply Robust Estimation for Local Distribution Shifts with Multiple Domains (ICML 2024 · similarity 0.87)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.87)
- Distributionally Robust Optimization with Bias and Variance Reduction (ICLR 2024 · similarity 0.86)
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.