Research map / Adversarial robustness and security / Certified and empirical robustness
A Practical Upper Bound for the Worst-Case Attribution Deviations
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 the proposed Taylor-based upper bound for attribution deviations under small perturbations | Gradient-based attribution methods · CIFAR-10, ℓ2 and ℓ∞ attacks | Whether attacked attribution distances remain within the bound | Helped |
| Use the generalized bound with an adaptive Taylor-remainder factor | ℓ2 attribution-deviation bounds · CIFAR-10 and other evaluated datasets and models | Coverage of attacked attribution distances: 100% | Helped |
| Use the ℓ∞ bound based on the absolute sum of the quadratic-form matrix instead of the ℓp relaxation | Attribution-deviation bounding · ℓ∞-norm perturbations | Tightness of the theoretical upper bound | Helped |
| Increase perturbation size ε (ablation) | Proposed ℓ2 upper bound with TRADES+IGR · CIFAR-10, label constraint | Euclidean attribution upper bound T_e: 0.1692 -> 0.3384 -> 0.8460 | Mixed |
Closest papers by meaning
- Rethinking Robustness of Model Attributions (AAAI 2024 · similarity 0.89)
- Enhanced Regularizers for Attributional Robustness (AAAI 2021 · similarity 0.89)
- Smoothed Geometry for Robust Attribution (NeurIPS 2020 · similarity 0.89)
- Exploiting the Relationship Between Kendall's Rank Correlation and Cosine Similarity for Attribution Protection (NeurIPS 2022 · similarity 0.88)
- Improving Adversarial Robustness of Attribution via Implicit Regularization (ICML 2026 · similarity 0.87)
- Adversarial Attacks on Data Attribution (ICLR 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.