Research map / Adversarial robustness and security / Certified and empirical robustness
Attributional Robustness Training using Input-Gradient Spatial Alignment
Mayank Singh, Nupur Kumari, Puneet Mangla, Abhishek Sinha, Vineeth N Balasubramanian, Balaji Krishnamurthy
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 ART instead of PGD-10 adversarial training | ResNet WRN-28-10 · CIFAR-10 | Top-k intersection: 69.00 -> 92.90 | Improved |
| use ART instead of PGD-7 adversarial training | WRN-40-2 · SVHN | Top-k intersection: 39.67 -> 61.37 | Improved |
| use ART instead of PGD-7 | WRN-28-10 · Flower | Top-k intersection: 80.84 -> 79.84 | Hurt |
| use ART instead of PGD-7 | WRN-28-10 · Flower | PGD-40 adversarial accuracy: 69.85 -> 33.08 | Hurt |
| use ART instead of PGD-10 | WRN-28-10 · CIFAR-10 | PGD-40 adversarial accuracy: 44.07 -> 37.58 | Hurt |
| replace cosine distance with l2 distance in L_attr (ablation) | ART · CIFAR-10 | Top-k intersection: 91.25 -> 68.41 | Hurt |
| add triplet loss with negative sample to cosine distance (ablation) | ART · CIFAR-10 | Top-k intersection: 91.25 -> 92.90 | Improved |
| choose negative as second most likely class instead of argmin logit (ablation) | ART · CIFAR-10 | Kendall correlation: 83.32 -> 91.76 | Improved |
| apply ART instead of PGD-10 for ResNet50 localization | ResNet50 · CUB | GT-Known Loc (Grad): 66.73 -> 82.65 | Improved |
| apply ART instead of natural training for localization | VGG-GAP · CUB | GT-Known Loc (Grad): 72.54 -> 76.50 | Improved |
| ART instead of PGD-10 under targeted attribution attack | ART · CIFAR-10 | Kendall tau: 36.29 -> 64.76 | Improved |
| raise lambda beyond 0.75 (ablation) | ART · CIFAR-10 | test accuracy | Hurt |
Closest papers by meaning
- Rethinking Robustness of Model Attributions (AAAI 2024 · similarity 0.90)
- Enhanced Regularizers for Attributional Robustness (AAAI 2021 · similarity 0.89)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 2025 · similarity 0.88)
- Probabilistic Stability Guarantees for Feature Attributions (NeurIPS 2025 · similarity 0.84)
- Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers (NeurIPS 2024 · similarity 0.84)
- Less is More: Fewer Interpretable Region via Submodular Subset Selection (ICLR 2024 · similarity 0.84)
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.
