Research map / Label-efficient and robust learning / Recognition and long-tail
CRAFT: Concept Recursive Activation FacTorization for Explainability
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 |
|---|---|---|---|
| provide CRAFT explanations instead of no explanation | CRAFT · Husky vs. Wolf | utility: 0.95 -> 1.53 | Helped |
| provide CRAFT explanations instead of no explanation | CRAFT · Leaves | utility: 1.02 -> 1.15 | Helped |
| provide CRAFT explanations instead of no explanation | CRAFT · "Kit Fox" vs. "Red Fox" | utility: 0.94 -> 0.87 | Hurt |
| add concept attribution maps to global concepts (ablation) | CRAFT · Husky vs. Wolf | utility: 1.38 -> 1.53 | Helped |
| add concept attribution maps to global concepts (ablation) | CRAFT · Leaves | utility: 1.12 -> 1.15 | Helped |
| add concept attribution maps to global concepts (ablation) | CRAFT · "Kit Fox" vs. "Red Fox" | utility: 0.88 -> 0.87 | Hurt |
| refine concepts recursively into sub-concepts (ablation) | CRAFT · human intruder detection, experts | accuracy: 70.19% -> 74.81% (p = 0.18) | No clear effect |
| refine concepts recursively into sub-concepts (ablation) | CRAFT · human intruder detection, laymen | accuracy: 61.08% -> 67.03% (p = 0.043) | Helped |
| choose a recursively refined sub-concept rather than its parent concept (ablation) | CRAFT · human binary choice, experts | sub-concept choice rate: 76.1% | Helped |
| choose a recursively refined sub-concept rather than its parent concept (ablation) | CRAFT · human binary choice, laymen | sub-concept choice rate: 74.95% | Helped |
| rank concepts with total Sobol indices instead of TCAV scores | CRAFT · concept-space fidelity analysis | deletion and insertion curves | Helped |
Closest papers by meaning
- Probabilistic Conceptual Explainers: Trustworthy Conceptual Explanations for Vision Foundation Models (ICML 2024 · similarity 0.88)
- CoE: Chain-of-Explanation via Automatic Visual Concept Circuit Description and Polysemanticity Quantification (CVPR 2025 · similarity 0.88)
- Saliency strikes back: How filtering out high frequencies improves white-box explanations (ICML 2024 · similarity 0.88)
- Unveiling Concept Attribution in Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.87)
- Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition (CVPR 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.