Research map / Probabilistic and scientific modeling / Molecules and proteins
Concept Activation Regions: A Generalized Framework For Concept-Based Explanations
Jonathan Crabbé, Mihaela van der Schaar
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 |
|---|---|---|---|
| Use TCAR scores instead of TCAV scores for global class-concept explanations | Concept-based explanations · MNIST | Pearson correlation with true concept proportion: .60 -> 1.00 | Improved |
| Use TCAR scores instead of TCAV scores for global class-concept explanations | Concept-based explanations · ECG | Pearson correlation with true concept proportion: .74 -> .99 | Improved |
| Use TCAR scores instead of TCAV scores for global class-concept explanations | Concept-based explanations · CUB | Pearson correlation with true concept proportion: .52 -> .86 | Improved |
| Tune kernel type, width, and error penalty with Bayesian optimization (ablation) | CAR classifier · CUB Inception penultimate layer | concept classifier accuracy: (93 ± .2) % -> (93 ± .2) % | No clear effect |
| Use CAR-based Integrated Gradients for concept density rather than vanilla Integrated Gradients for model output | Feature importance · MNIST, ECG, and CUB | Pearson correlation with vanilla feature importance | Mixed |
| Use CAR-based feature importance to compare related concepts | Concept feature importance · MNIST | Pearson correlation between Loop and Curvature importance: r=.71 | Improved |
| Use CAR-based feature importance to compare mutually exclusive concepts | Concept feature importance · CUB | Pearson correlation between striped and solid back pattern importance: r=-.86 | Mixed |
| Use the full concept density rather than a sparse density using only SVC support vectors (ablation) | CAR-based feature importance · Concept feature importance | feature attribution | Improved |
| Increase concept-set size beyond N^c > 200 (ablation) | CAR classifier · MNIST | concept classifier accuracy | No clear effect |
Closest papers by meaning
- CE-FAM: Concept-Based Explanation via Fusion of Activation Maps (ICCV 2025 · similarity 0.91)
- Local Concept Embeddings for Analysis of Concept Distributions in Vision DNN Feature Spaces (IJCV 2025 · similarity 0.90)
- FastCAV: Efficient Computation of Concept Activation Vectors for Explaining Deep Neural Networks (ICML 2025 · similarity 0.89)
- FaCT: Faithful Concept Traces for Explaining Neural Network Decisions (NeurIPS 2025 · similarity 0.88)
- Navigating Neural Space: Revisiting Concept Activation Vectors to Overcome Directional Divergence (ICLR 2025 · similarity 0.88)
- Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models (CVPR 2024 · similarity 0.86)
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.
