Research map / Label-efficient and robust learning / Recognition and long-tail
The effectiveness of feature attribution methods and its correlation with automatic evaluation scores
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 |
|---|---|---|---|
| show confidence score plus heatmap to humans instead of AI alone | ResNet-34 human-AI team · ImageNet natural | team accuracy (%): 80.79 -> 88.77 | Helped |
| add heatmap explanation on top of confidence score | human-AI team · ImageNet natural | team accuracy (%): 84.79 -> 88.77 | Helped |
| human with confidence score or heatmap instead of AI alone | human-AI team · Stanford Dogs natural | team accuracy (%): 81.14 -> 76.45 | Hurt |
| use 3-NN examples instead of AI alone | human-AI team · Stanford Dogs natural | team accuracy (%): 81.14 -> 82.88 | Helped |
| replace GradCAM attribution maps with 3-NN examples for experts | human-AI team · ImageNet expert study | mean per-user accuracy (%): 68.00 -> 76.67 | Helped |
| use 3-NN or EP explanations instead of confidence only | human-AI team · Stanford Dogs (real and adversarial) | user accuracy | Hurt |
| use 3-NN instead of GradCAM/SOD | human-AI team · ImageNet lay users | per-user accuracy: ~3% margin | Helped |
| use GradCAM vs EP vs SOD heatmaps | human-AI team · ImageNet | per-user accuracy | No clear effect |
| higher automatic localization scores (IoU, WSL, Pointing Game) of attribution maps | GradCAM and EP · human-AI team accuracy | Pearson correlation: r=0.22 IoU on ImageNet GradCAM; other cases -0.12 <= r <= 0.07 | No clear effect |
Closest papers by meaning
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.90)
- quanda: An Interpretability Toolkit for Training Data Attribution Evaluation (NeurIPS 2026 · similarity 0.88)
- Rethinking Robustness of Model Attributions (AAAI 2024 · similarity 0.88)
- $\texttt{dattri}$: A Library for Efficient Data Attribution (NeurIPS 2024 · similarity 0.86)
- One Wave To Explain Them All: A Unifying Perspective On Feature Attribution (ICML 2025 · similarity 0.86)
- Human and AI Perceptual Differences in Image Classification Errors (AAAI 2025 · 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.