Research map / Architectures and efficiency / RNNs and normalization
On Spectral Properties of Gradient-based Explanation Methods
Amir Mehrpanah, Erik Englesson, Hossein Azizpour
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 squared gradients instead of gradients in SmoothGrad (ablation) | SmoothGrad · ImageNet | Insertion: 0.212 -> 0.465 | Improved |
| use squared gradients instead of gradients in Integrated Gradients (ablation) | IG · ImageNet | Insertion: 0.201 -> 0.340 | Improved |
| use squared gradients instead of gradients in SmoothGrad (ablation) | SmoothGrad · Food101 | Deletion: 0.137 -> 0.162 | Hurt |
| use squared gradients instead of gradients in XIG (ablation) | XIG · Food101 | Deletion: 0.177 -> 0.263 | Hurt |
| use per-image optimal perturbation scale (max cosine similarity) in SG^2 (ablation) | SG^2 · ImageNet | Insertion: 0.465 -> 0.491 | Improved |
| use per-image optimal perturbation scale in SG^2 (ablation) | SG^2 · Food101 | Insertion: 0.371 -> 0.394 | Improved |
| use per-image optimal perturbation scale in SG^2 (ablation) | SG^2 · ImageNet | Deletion: 0.157 -> 0.187 | Hurt |
| aggregate over noise scales with uniform explanation prior (SpectralLens) instead of Dirac prior (ablation) | SG^2 · ImageNet | Insertion: 0.465 -> 0.493 | Improved |
| aggregate over noise scales with uniform explanation prior (SpectralLens) (ablation) | SG^2 · Food101 | Deletion: 0.162 -> 0.157 | Improved |
| apply input multiplication (XIG) instead of IG (ablation) | Integrated Gradients · Food101 | Insertion: 0.145 -> 0.101 | Hurt |
Closest papers by meaning
- On the Complexity-Faithfulness Trade-off of Gradient-Based Explanations (ICCV 2025 · similarity 0.90)
- Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanations (ICML 2025 · similarity 0.87)
- Towards Modeling Uncertainties of Self-Explaining Neural Networks via Conformal Prediction (AAAI 2024 · similarity 0.86)
- GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking (ICLR 2024 · similarity 0.86)
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.86)
- SHAP Meets Tensor Networks: Provably Tractable Explanations with Parallelism (NeurIPS 2025 · 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.
