Research map / Label-efficient and robust learning / OOD and anomaly detection
Class Is Invariant to Context and Vice Versa: On Learning Invariance for Out-of-Distribution Generalization
Jiaxin Qi, Kaihua Tang, Qianru Sun, Xian-Sheng Hua, Hanwang Zhang
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 |
|---|---|---|---|
| estimate context with IRMCon features and reweight (vs LfF reweighting) | ERM-IPW classifier · Colored MNIST (99.9% bias) | accuracy (%): 56.8±1.6 -> 66.7±2.3 | Improved |
| estimate context with IRMCon features and reweight (vs LfF reweighting) | ERM-IPW classifier · Colored MNIST (99.5% bias) | accuracy (%): 77.0±1.5 -> 81.0±0.9 | Improved |
| estimate context with IRMCon features and reweight (vs LfF) | ERM-IPW classifier · Colored MNIST (95.0% bias) | accuracy (%): 86.8±0.5 -> 92.2±0.5 | Improved |
| use IRMCon-IPW instead of ERM | ResNet-18 classifier · Colored MNIST (99.9% bias) | accuracy (%): 20.4±1.1 -> 66.7±2.3 | Improved |
| use IRMCon-IPW instead of ERM | ResNet-18 classifier · Corrupted CIFAR-10 (95.0% bias) | accuracy (%): 39.9±1.6 -> 53.8±1.3 | Improved |
| use IRMCon-IPW instead of ERM | ResNet-18 classifier · BAR (99.0% bias) | accuracy (%): 52.9±0.7 -> 55.3±0.6 | Improved |
| use IRMCon-IPW instead of Feat-Aug | reweighting classifier · Colored MNIST (95.0% bias) | accuracy (%): 89.7±1.1 -> 92.2±0.5 | Improved |
| use LfF reweighting instead of plain ERM | ResNet-18 · BAR (99.0% bias) | accuracy (%): 52.9±0.7 -> 48.1±2.7 | Hurt |
| use LfF reweighting instead of plain ERM | ResNet-18 · PACS average (from scratch) | accuracy (%): 54.3 -> 51.8 | Hurt |
| use IRMCon-IPW instead of ERM | ResNet-18 · PACS average (from scratch) | accuracy (%): 54.3 -> 55.7 | Improved |
| use IRMCon-IPW instead of Fish (domain labels) | ResNet-18 · PACS average (from scratch) | accuracy (%): 56.6 -> 55.7 | No clear effect |
| apply traditional reweighting on a context-balanced training set | LfF reweighting · Colored MNIST (context-balanced) | accuracy (%) | Hurt |
Closest papers by meaning
- Rethinking Invariance in In-context Learning (ICLR 2025 · similarity 0.86)
- Diagnosing and Rectifying Fake OOD Invariance: A Restructured Causal Approach (AAAI 2024 · similarity 0.86)
- HYPO: Hyperspherical Out-Of-Distribution Generalization (ICLR 2024 · similarity 0.85)
- In-Context Learning and Occam's Razor (ICML 2025 · similarity 0.85)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · similarity 0.85)
- Aggregation Hides Out-of-Distribution Generalization Failures from Spurious Correlations (NeurIPS 2025 · 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.
