Research map / Vision-language and multimodal / Human and embodied modeling
Prototype-Grounded Concept Models for Verifiable Concept Alignment
Stefano Colamonaco, David Debot, Pietro Barbiero, Giuseppe Marra
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 |
|---|---|---|---|
| remove wrong prototypes after training on noisy concept labels (ablation) | PGCM · ColorMNIST+ (noisy labels) | concept accuracy: 92.9 -> 96.9 | Improved |
| edit concepts of wrong prototypes after training on noisy concept labels (ablation) | PGCM · ColorMNIST+ (noisy labels) | concept accuracy: 92.8 -> 97.8 | Improved |
| remove wrong prototypes after training on noisy concept labels (ablation) | PGCM · CLEVR-Hans3 (noisy labels) | concept accuracy: 81.6 -> 95.1 | Improved |
| edit concepts of wrong prototypes after training on noisy concept labels (ablation) | PGCM · CLEVR-Hans3 (noisy labels) | concept accuracy: 81.6 -> 97.5 | Improved |
| increase the number of learned prototypes (ablation) | PGCM · CLEVR-Hans3 | concept and task accuracy | Improved |
| ground concepts in prototypes instead of a standard CBM | PGCM vs CBM · CelebA | concept accuracy: 81.3 -> 78.5 | Hurt |
| ground concepts in prototypes instead of a standard CBM | PGCM vs CBM · CelebA | task accuracy: 84.0 -> 83.0 | Hurt |
| ground concepts in prototypes instead of a standard CBM | PGCM vs CBM · ColorMNIST+ and CUB | task accuracy: 99.6 -> 99.7 (ColorMNIST+); 63.6 -> 63.4 (CUB) | No clear effect |
| use prototype-aware concept interventions that update multiple concepts at once | PGCM · ColorMNIST+ and CUB | concept accuracy after interventions | Improved |
| use prototype-aware concept interventions instead of standard ones | PGCM · CelebA | concept accuracy after interventions | Hurt |
Closest papers by meaning
- VLG-CBM: Training Concept Bottleneck Models with Vision-Language Guidance (NeurIPS 2024 · similarity 0.89)
- Hybrid Concept Bottleneck Models (CVPR 2025 · similarity 0.88)
- Coarse-to-Fine Concept Bottleneck Models (NeurIPS 2024 · similarity 0.88)
- Energy-Based Concept Bottleneck Models: Unifying Prediction, Concept Intervention, and Probabilistic Interpretations (ICLR 2024 · similarity 0.88)
- Concepts Worth Having: Refining VLM-Guided Concept Bottleneck Models with Minimal Annotations (NeurIPS 2026 · similarity 0.88)
- DCBM: Data-Efficient Visual Concept Bottleneck Models (ICML 2025 · similarity 0.87)
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.
