Research map / Learning theory and algorithms / Prediction and decision losses
Towards the Unification and Robustness of Perturbation and Gradient Based Explanations
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 |
|---|---|---|---|
| Increase the number of perturbations | SmoothGrad and C-LIME · Simulated, Bankruptcy, and Online Shopping | L1 distance between explanations: 100 perturbations | Helped |
| Increase the number of perturbations | SmoothGrad and C-LIME · Simulated, Bankruptcy, and Online Shopping | average maximum L1 distance between explanations for nearby inputs: 200 perturbations | Helped |
| Use SmoothGrad instead of C-LIME | SmoothGrad and C-LIME · Simulated, Bankruptcy, and Online Shopping | average maximum L1 distance between explanations for nearby inputs | Helped |
| Use SmoothGrad instead of C-LIME | SmoothGrad and C-LIME · Simulated, Bankruptcy, and Online Shopping | perturbations required for robustness saturation: 200 | Helped |
| Reduce perturbation variance (ablation) | SmoothGrad and C-LIME · Bankruptcy | perturbations required for equivalence: 0.01, 0.1, and 1 | Helped |
| Reduce perturbation variance (ablation) | SmoothGrad and C-LIME · Bankruptcy | perturbations required for robustness: 0.01, 0.1, and 1 | Helped |
| Reduce the underlying model's accuracy (ablation) | SmoothGrad and C-LIME · Bankruptcy | convergence rate: 86% accuracy using 16 epochs; 96% accuracy after 60 epochs | Hurt |
Closest papers by meaning
- Improving Perturbation-based Explanations by Understanding the Role of Uncertainty Calibration (NeurIPS 2025 · similarity 0.88)
- Robust and Stable Black Box Explanations (ICML 2020 · similarity 0.88)
- Use perturbations when learning from explanations (NeurIPS 2023 · similarity 0.87)
- Locally Invariant Explanations: Towards Stable and Unidirectional Explanations through Local Invariant Learning (NeurIPS 2023 · similarity 0.87)
- Constraint-Driven Explanations for Black-Box ML Models (AAAI 2022 · similarity 0.87)
- 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.