Research map / Label-efficient and robust learning / Recognition and long-tail
Constructing Concept-based Models to Mitigate Spurious Correlations with Minimal Human Effort
Jeeyung Kim, Ze Wang, Qiang Qiu
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 |
|---|---|---|---|
| add automatic annotation refinement (VFM chain) for two LLaVA-annotated attributes (ablation) | LLaVA-based CBM · ImageNet-Opener | worst-group accuracy: 80.0% -> 90.0% | Improved |
| add annotation refinement (ablation) | LLaVA-based CBM · ImageNet-Opener | average accuracy: 82.4% -> 86.1% | Improved |
| annotate concepts with LLaVA instead of CLIP (same CUB concept set) (ablation) | CBM · Waterbirds | worst-group accuracy: 31.6% -> 80.8% (112 CUB attributes) | Improved |
| annotate concepts with LLaVA instead of CLIP (Label-free concept set) (ablation) | CBM · Waterbirds | worst-group accuracy: 55.0% -> 80.5% | Improved |
| use LLaVA-based CBM instead of ERM | ResNet-50 · Metashifts | worst-group accuracy: 62.1% -> 78.0% | Improved |
| use LLaVA-based CBM instead of Group DRO oracle | ResNet-50 · Waterbirds | worst-group accuracy: 91.4% -> 83.2% | Hurt |
| use LLaVA-based CBM instead of ERM | ResNet-50 · Waterbirds | average accuracy: 97.3% -> 94.2% | Hurt |
| filter spurious-correlated concepts (point-biserial correlation > 0.2) from the concept pool | LLaVA-based CBM · ImageNet-Opener, Metashifts | concept pool size: 47 -> 42 concepts (ImageNet-Opener); 28 -> 24 (Metashifts) | Improved |
| refine wood material annotation with VFM chain (ablation) | LLaVA annotations · ImageNet-Opener can opener | share labeled wood: 21% -> 3% | Improved |
Closest papers by meaning
- Stochastic Concept Bottleneck Models (NeurIPS 2024 · similarity 0.92)
- An Analysis of Concept Bottleneck Models: Measuring, Understanding, and Mitigating the Impact of Noisy Annotations (NeurIPS 2025 · similarity 0.91)
- Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions (NeurIPS 2025 · similarity 0.91)
- Bayesian Concept Bottleneck Models with LLM Priors (NeurIPS 2025 · similarity 0.90)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.90)
- Hybrid Concept Bottleneck Models (CVPR 2025 · similarity 0.90)
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.
