Research map / Architectures and efficiency / CNN design
Invertible Concept-based Explanations for CNN Models with Non-negative Concept Activation Vectors
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 NMF rather than PCA for concept vectors | ICE concept explanations · ILSVRC2012, ResNet50 (RI) | task prediction accuracy: 37.8% ± 5.9% -> 74.4% ± 9.2% | Helped |
| Use NMF rather than PCA for concept vectors | ICE concept explanations · ILSVRC2012, Inception-V3 (II) | task prediction accuracy: 40.0% ± 8.6% -> 62.6% ± 18.6% | Helped |
| Use NMF rather than PCA for concept vectors | ICE concept explanations · CUB, ResNet50 (RC) | task prediction accuracy: 57.0% ± 11.6% -> 81.1% ± 8.4% | Helped |
| Use NMF rather than clustering for concept vectors | ICE concept explanations · ILSVRC2012, Inception-V3 (II) | task prediction accuracy: 44.8% ± 13.2% -> 62.6% ± 18.6% | Helped |
| Use NMF rather than clustering for concept vectors | ICE concept explanations · ILSVRC2012, ResNet50 (RI) | task prediction accuracy: 66.3% ± 13.8% -> 74.4% ± 9.2% | No clear effect |
| Use NMF rather than clustering for concept vectors | ICE concept explanations · CUB, ResNet50 (RC) | task prediction accuracy: 78.6% ± 15.5% -> 81.1% ± 8.4% | No clear effect |
| Use NMF rather than PCA for concept vectors | ICE concept explanations · ILSVRC2012, ResNet50 (RI) | description similarity: 0.52 ± 0.08 -> 0.59 ± 0.1 | No clear effect |
| Use NMF rather than PCA for concept vectors | ICE concept explanations · ILSVRC2012, Inception-V3 (II) | description similarity: 0.49 ± 0.08 -> 0.57 ± 0.08 | No clear effect |
| Use NMF rather than PCA for concept vectors | ICE concept explanations · CUB, ResNet50 (RC) | description similarity: 0.59 ± 0.03 -> 0.7 ± 0.04 | Helped |
| Use PCA rather than NMF for the reducer | ICE approximate models · ImageNet and CUB fidelity experiments | fidelity | Helped |
Closest papers by meaning
- FastCAV: Efficient Computation of Concept Activation Vectors for Explaining Deep Neural Networks (ICML 2025 · similarity 0.90)
- Navigating Neural Space: Revisiting Concept Activation Vectors to Overcome Directional Divergence (ICLR 2025 · similarity 0.88)
- Show and Tell: Visually Explainable Deep Neural Nets via Spatially-Aware Concept Bottleneck Models (CVPR 2025 · similarity 0.87)
- FACE: Faithful Automatic Concept Extraction (NeurIPS 2025 · similarity 0.87)
- Towards Human-Understandable Multi-Dimensional Concept Discovery (CVPR 2025 · similarity 0.87)
- Enhancing Performance of Explainable AI Models with Constrained Concept Refinement (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.