Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Automatic Concept-based Explanations
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 |
|---|---|---|---|
| Use automatically discovered ACE concepts instead of hand-labeled concepts | ImageNet concept explanations · Intruder detection | human accuracy: 97% (14.6/15) (± 0.7) -> 99% (14.9/15) (± 0.3) | No clear effect |
| Present grouped ACE segments rather than random segments | ACE · ImageNet concept meaningfulness choice | human selection rate: 95.6% (14.3/15) (± 1.0) | Helped |
| Add the five most important concepts | Inception-V3 predictions on ImageNet validation images · ImageNet validation images, 100 classes | prediction accuracy relative to original: within 80% of the original accuracy | Helped |
| Remove the five most important concepts | Inception-V3 predictions on ImageNet validation images · ImageNet validation images, 100 classes | misclassification rate among originally correct samples: more than 80% | Hurt |
| Stitch the four most important concepts onto blank images | Inception-V3 · ImageNet stitched images | classification accuracy: 20 classes (zebra, liner, etc), more than 80%; for more than half of the classes, above 40% | Helped |
| Use K-Means clustering rather than the tested alternative clustering methods (ablation) | ACE concept grouping · ImageNet concept discovery | qualitative clustering results | Helped |
Closest papers by meaning
- Towards Compositionality in Concept Learning (ICML 2024 · similarity 0.86)
- CoE: Chain-of-Explanation via Automatic Visual Concept Circuit Description and Polysemanticity Quantification (CVPR 2025 · similarity 0.86)
- Understanding Inter-Concept Relationships in Concept-Based Models (ICML 2024 · similarity 0.85)
- Towards Human-Understandable Multi-Dimensional Concept Discovery (CVPR 2025 · similarity 0.85)
- MCPNet: An Interpretable Classifier via Multi-Level Concept Prototypes (CVPR 2024 · similarity 0.85)
- Restyling Unsupervised Concept Based Interpretable Networks with Generative Models (ICLR 2025 · similarity 0.85)
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.