Research map / Adversarial robustness and security / Certified and empirical robustness
Smoothed Geometry for Robust Attribution
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 |
|---|---|---|---|
| Apply Gaussian stochastic smoothing with Smooth Gradient instead of Saliency Map | ResNet-20 attribution on CIFAR-10 · CIFAR-10, TopK Attack | cosd AUC: 0.70 -> 0.62 | Helped |
| Apply uniform stochastic smoothing with Uniform Gradient instead of Saliency Map | ResNet-20 attribution on CIFAR-10 · CIFAR-10, TopK Attack | cosd AUC: 0.70 -> 0.67 | Helped |
| Apply Gaussian stochastic smoothing with Smooth Gradient instead of Saliency Map | ResNet-20 attribution on CIFAR-10 · CIFAR-10, Manipulate Attack | cosd AUC: 0.72 -> 0.59 | Helped |
| Apply uniform stochastic smoothing with Uniform Gradient instead of Saliency Map | ResNet-20 attribution on CIFAR-10 · CIFAR-10, Manipulate Attack | cosd AUC: 0.72 -> 0.54 | Helped |
| Use Uniform Gradient rather than Smooth Gradient | ResNet-20 attribution on CIFAR-10 · CIFAR-10, TopK Attack | cosd AUC: 0.62 -> 0.67 | Hurt |
| Use Uniform Gradient rather than Smooth Gradient | ResNet-20 attribution on CIFAR-10 · CIFAR-10, Manipulate Attack | cosd AUC: 0.59 -> 0.54 | Helped |
| Smooth the model geometry using Smooth Surface Regularization during training | Models evaluated with gradient-based attribution attacks · CIFAR-10 | attribution robustness | Helped |
| Use adversarial training rather than natural training | Models evaluated with gradient-based attribution attacks · CIFAR-10 | attribution robustness | Helped |
| Transfer a Saliency Map-targeted perturbation to Integrated Gradients rather than a smoothed attribution | ResNet-50 attribution maps · ImageNet | change in attribution map | Mixed |
| Use Smooth Gradient or Uniform Gradient instead of Integrated Gradients under a Saliency Map-targeted transfer attack | ResNet-50 attribution maps · ImageNet | change in attribution map | Helped |
Closest papers by meaning
- Enhanced Regularizers for Attributional Robustness (AAAI 2021 · similarity 0.90)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 2025 · similarity 0.89)
- Improving Adversarial Robustness of Attribution via Implicit Regularization (ICML 2026 · similarity 0.89)
- Rethinking Robustness of Model Attributions (AAAI 2024 · similarity 0.88)
- Manifold Integrated Gradients: Riemannian Geometry for Feature Attribution (ICML 2024 · similarity 0.87)
- Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness (NeurIPS 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.