Research map / Label-efficient and robust learning / OOD and anomaly detection
DecAug: Out-of-Distribution Generalization via Decomposed Feature Representation and Semantic Augmentation
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 |
|---|---|---|---|
| apply DecAug rather than the strongest reported baseline | PACS classification with ResNet-18 · PACS | accuracy (average): 81.60 -> 82.39 | Helped |
| apply DecAug rather than the strongest reported baseline | Colored MNIST test environment · Colored MNIST test env | accuracy: 68.70 $\pm$ 0.9 -> 69.60 $\pm$ 2.0 | Helped |
| apply DecAug rather than JiGen | NICO Animal classification · NICO Animal | accuracy: 84.95 -> 85.23 | Helped |
| apply DecAug rather than JiGen | NICO Vehicle classification · NICO Vehicle | accuracy: 79.45 -> 80.12 | Helped |
| add orthogonal gradient loss (ablation) | DecAug on PACS · PACS | accuracy (average): 80.77 -> 82.39 | Helped |
| set orthogonal regularization coefficient to 0.0005 or 0.001 instead of 0.01 (ablation) | DecAug on PACS · PACS | accuracy (average): 80.76; 81.83; 82.39 | Hurt |
| replace gradient orthogonalization with orthogonality between features (ablation) | DecAug on PACS · PACS | accuracy (average): 79.90 -> 82.39 | Hurt |
| replace gradient orthogonalization with DANN loss (ablation) | DecAug on PACS · PACS | accuracy (average): 81.00 -> 82.39 | Hurt |
| add confusion regularization (ablation) | DecAug · PACS | accuracy | No clear effect |
| add adversarial augmentation to Jigsaw instead of augmenting the context-predicting branch in DecAug (ablation) | Jigsaw on PACS · PACS | accuracy | Hurt |
| add semantic augmentation to context features (ablation) | DecAug on PACS · PACS | accuracy (average): 79.64 -> 82.39 | Helped |
| vary semantic augmentation magnitude ε (ablation) | DecAug on PACS · PACS | accuracy (average): 79.36; 80.55; 73.07; 61.86; 82.39 | Mixed |
Closest papers by meaning
- Intermediate Layer Classifiers for OOD generalization (ICLR 2025 · similarity 0.85)
- HYPO: Hyperspherical Out-Of-Distribution Generalization (ICLR 2024 · similarity 0.85)
- AHA: Human-Assisted Out-of-Distribution Generalization and Detection (NeurIPS 2024 · similarity 0.85)
- Minimal Semantic Sufficiency Meets Unsupervised Domain Generalization (NeurIPS 2025 · similarity 0.85)
- Overcoming the Pitfalls of Vision-Language Model Finetuning for OOD Generalization (ICLR 2024 · similarity 0.84)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · 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.