Research map / Learning theory and algorithms / Sample complexity
(Almost) Provable Error Bounds Under Distribution Shift via Disagreement Discrepancy
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 logits instead of full features for the critic (ablation) | Dis² · All shifts, non-domain-adversarial | MAE: 0.2807 -> 0.1504 | Helped |
| Use logits instead of full features for the critic (ablation) | Dis² · All shifts, non-domain-adversarial | coverage: 1.0000 -> 0.9889 | No clear effect |
| Use logits instead of full features for the critic (ablation) | Dis² · All shifts, domain-adversarial | MAE: 0.1918 -> 0.0935 | Helped |
| Use logits instead of full features for the critic (ablation) | Dis² · All shifts, domain-adversarial | coverage: 1.0000 -> 0.7500 | Hurt |
| Drop the concentration term (ablation) | Dis² (logits) · All shifts, non-domain-adversarial | MAE: 0.1504 -> 0.0829 | Helped |
| Drop the concentration term (ablation) | Dis² (logits) · All shifts, non-domain-adversarial | coverage: 0.9889 -> 0.7556 | Hurt |
| Drop the concentration term (ablation) | Dis² (logits) · All shifts, domain-adversarial | MAE: 0.0935 -> 0.0639 | Helped |
| Drop the concentration term (ablation) | Dis² (logits) · All shifts, domain-adversarial | coverage: 0.7500 -> 0.4167 | Hurt |
| Use the proposed disagreement logistic loss instead of the alternative disagreement losses (ablation) | Dis² critic optimization · Across evaluated shifts | disagreement discrepancy: + statistically significant increases; values not reported in supplied text | Helped |
| Optimize the critic on fewer principal components (ablation) | Dis² feature representation · Distribution-shift benchmarks | error-bound validity | Mixed |
| Use logits rather than progressively fewer top principal components (ablation) | Dis² critic representation · Distribution-shift benchmarks | error-bound validity | Helped |
| Add the concentration term to the sample-average estimate (ablation) | Dis² without δ · All shifts | coverage: 0.7556 -> 0.9889; 0.4167 -> 0.7500 | Helped |
Closest papers by meaning
- Exact Upper and Lower Bounds for the Output Distribution of Neural Networks with Random Inputs (ICML 2025 · similarity 0.86)
- Learning Neural Networks with Distribution Shift: Efficiently Certifiable Guarantees (ICLR 2025 · similarity 0.86)
- Sharper Guarantees for Learning Neural Network Classifiers with Gradient Methods (ICLR 2025 · similarity 0.86)
- Tolerant Algorithms for Learning with Arbitrary Covariate Shift (NeurIPS 2024 · similarity 0.86)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.86)
- A Universal Growth Rate for Learning with Smooth Surrogate Losses (NeurIPS 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.