Research map / Learning theory and algorithms / Prediction and decision losses
Evaluating the Robustness of Interpretability Methods through Explanation Invariance and Equivariance
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 an invariant baseline with feature-importance methods | Gradient-based and perturbation-based feature importance · ECG and FashionMNIST | explanation equivariance | Helped |
| Use loss-based rather than representation-based example importance | Example-importance methods · Experiments across datasets | explanation invariance | Helped |
| Use an invariant model layer for representation-based example importance | Representation-based example-importance methods · Experiments across datasets | explanation invariance | Helped |
| Use an invariant model layer for concept classifiers | Concept-based explanations · Experiments across datasets | explanation invariance | Helped |
| Aggregate explanations over sampled input symmetries | Concept-importance methods · ECG and FashionMNIST | explanation invariance | Helped |
| Relax the explained model's symmetry invariance | Interpretability methods, especially feature-importance methods · CIFAR100, STL10, ECG, and FashionMNIST | explanation invariance or equivariance | Hurt |
| Use an approximately invariant model instead of a perfectly invariant model | Feature-importance methods · ECG and FashionMNIST | explanation equivariance | Hurt |
| Use concept importance on an equivariant rather than invariant layer | CAV · ModelNet40 | explanation invariance | Hurt |
Closest papers by meaning
- Beyond Interpretability: The Gains of Feature Monosemanticity on Model Robustness (ICLR 2025 · similarity 0.87)
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.87)
- Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable? (ICLR 2025 · similarity 0.87)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.87)
- Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift (NeurIPS 2026 · similarity 0.87)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (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.