Research map / Label-efficient and robust learning / Recognition and long-tail
SLIM: Spuriousness Mitigation with Minimal Human Annotations
Xiwei Xuan, Ziquan Deng, Hsuan-Tien Lin, Kwan-Liu Ma
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 |
|---|---|---|---|
| curate a feature-balanced subset with SLIM instead of plain ERM | ResNet50 · Waterbirds | worst-group accuracy (%): 62.6 -> 89.1 (SLIM_Tr), 91.0 (SLIM_Val) | Improved |
| use SLIM_Tr with attention labels instead of CNC with extra training | ResNet50 · ImageNet-9 | worst-group accuracy (%): 58.9 -> 68.1 | Improved |
| use SLIM_Tr instead of GDRO with full spuriousness labels | ResNet50 · CelebA | worst-group accuracy (%): 88.9 -> 89.4 | No clear effect |
| use attention-correctness labels instead of spuriousness labels | human annotation · NICO labeling study | annotation consistency (%): 70.1 -> 95.1 | Improved |
| increase number of attention annotations (ablation) | SLIM_Tr · Waterbirds | worst-group accuracy (%): 78.2 -> 89.1 (60 -> 120 annotations) | Improved |
| increase annotations beyond 120 (ablation) | SLIM_Tr · Waterbirds | worst-group accuracy (%): 89.1 -> 89.4 (120 -> 150) | No clear effect |
| enlarge constructed training subset beyond the chosen size (ablation) | SLIM_Tr · Waterbirds | worst-group accuracy (%): 89.1 -> 83.5 (10% -> 15%) | Hurt |
| use attention space (attention-weighted features) instead of original feature space (ablation) | SLIM label propagation · Waterbirds | neighbor attention consistency | Improved |
| apply SLIM_Tr on ISIC with attention labels | ResNet50 · ISIC | worst-group accuracy (%): 56.7 -> 79.1 | Improved |
| apply SLIM to improve attention accuracy | ResNet50 · Waterbirds | worst-group AIoU: .25 -> .53 | Improved |
Closest papers by meaning
- Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples (CVPR 2025 · similarity 0.89)
- Severing Spurious Correlations with Data Pruning (ICLR 2025 · similarity 0.87)
- Escaping the SpuriVerse: Can Large Vision-Language Models Generalize Beyond Seen Spurious Correlations? (NeurIPS 2025 · similarity 0.85)
- Complexity Matters: Feature Learning in the Presence of Spurious Correlations (ICML 2024 · similarity 0.84)
- Views Can Be Deceiving: Improved SSL Through Feature Space Augmentation (ICLR 2024 · similarity 0.84)
- How Spurious Features are Memorized: Precise Analysis for Random and NTK Features (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.
