Research map / Learning theory and algorithms / Prediction and decision losses
On the (In)fidelity and Sensitivity of Explanations
Chih-Kuan Yeh, Cheng-Yu Hsieh, Arun Suggala, David I Inouye, Pradeep K Ravikumar
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 |
|---|---|---|---|
| Apply SmoothGrad to Grad (ablation) | Grad · MNIST global | sensitivity: 0.56 -> 0.28 | Improved |
| Apply SmoothGrad to Grad (ablation) | Grad · MNIST global | infidelity: 2.38 -> 1.89 | Improved |
| Apply SmoothGrad to base explanations (ablation) | Grad, IG, and GBP · MNIST, CIFAR-10, and ImageNet global | sensitivity and infidelity | Improved |
| Use the optimal explanation for square-removal perturbations | Global explanations · MNIST, CIFAR-10, and ImageNet | infidelity | Improved |
| Use the optimal explanation for noisy-baseline perturbations | Local explanations · MNIST | infidelity | Improved |
| Apply the optimal infidelity explanation | Explanations in the human evaluation · Human evaluation | human accuracy: 0.53 -> 0.88 | Improved |
| Increase SmoothGrad smoothing radius (ablation) | Grad-SG · MNIST local | infidelity | Mixed |
| Increase SmoothGrad smoothing radius (ablation) | Grad-SG · MNIST local | sensitivity | Improved |
| Adversarially train the predictor (ablation) | Gradient-based explanations · MNIST robust vs regular network | sensitivity and infidelity | Improved |
| Use a constant explanation | Any explanation · All inputs | sensitivity | Improved |
Closest papers by meaning
- Selective Explanations (NeurIPS 2024 · similarity 0.86)
- Saliency strikes back: How filtering out high frequencies improves white-box explanations (ICML 2024 · similarity 0.86)
- On the Faithfulness of Vision Transformer Explanations (CVPR 2024 · similarity 0.85)
- Efficient and Accurate Explanation Estimation with Distribution Compression (ICLR 2025 · similarity 0.85)
- Improving Perturbation-based Explanations by Understanding the Role of Uncertainty Calibration (NeurIPS 2025 · similarity 0.85)
- Trust Regions for Explanations via Black-Box Probabilistic Certification (ICML 2024 · 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.
