Research map / Adversarial robustness and security / Attacks on text and graphs
Enhanced Regularizers for Attributional Robustness
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 |
|---|---|---|---|
| add CACR and WACR | Attributionally robust training · Flower, Fashion-MNIST, MNIST, and GTSRB | Top-K and Kendall: Flower Top-K: 66.33% -> 69.50%; Kendall: 0.7974 -> 0.8121 | Helped |
| add CACR and WACR | Attributionally robust training with WRN 28-10 · Flower | Top-K and Kendall: Top-K: 79.84% -> 95.50%; Kendall: 0.8487 -> 0.9770 | Helped |
| add CACR and WACR | Attributionally robust training with WRN 28-10 · GTSRB | Top-K and Kendall: Top-K: 91.96% -> 92.14%; Kendall: 0.8934 -> 0.9502 | Helped |
| remove WACR (ablation) | Full method · Flower | Top-K and Kendall: Top-K: 69.50% -> 67.95%; Kendall: 0.8121 -> 0.8012 | Hurt |
| remove CACR (ablation) | Full method · Flower | Top-K and Kendall: Top-K: 69.50% -> 68.41%; Kendall: 0.8121 -> 0.8065 | Hurt |
| remove WACR (ablation) | Full method · Fashion-MNIST | Top-K: 80.19% -> 78.77% | Hurt |
| remove CACR (ablation) | Full method · Fashion-MNIST | Top-K: 80.19% -> 79.11% | Hurt |
| remove WACR (ablation) | Full method · MNIST | Top-K: 79.18% -> 76.62% | Hurt |
| remove CACR (ablation) | Full method · MNIST | Top-K: 79.18% -> 77.88% | Hurt |
| reduce WACR coefficient from 1 to 0.7 (ablation) | Full method · Flower, Fashion-MNIST, and MNIST | Top-K: Flower: 69.50 -> 69.04; Fashion-MNIST: 80.19% -> 80.19%; MNIST: 79.18% -> 79.18% | Hurt |
| reduce CACR coefficient from 1 to 0.7 (ablation) | Full method · Flower, Fashion-MNIST, and MNIST | Top-K: Flower: 69.50 -> 68.63%; Fashion-MNIST: 80.19% -> 79.86%; MNIST: 79.18% -> 78.83% | Hurt |
| remove KL(P_ME || P_NCA) from CACR (ablation) | Full method · Attribution maps | focus on true-class pixels | Hurt |
Closest papers by meaning
- Improving Adversarial Robustness of Attribution via Implicit Regularization (ICML 2026 · similarity 0.91)
- Robust Attribution Regularization (NeurIPS 2019 · similarity 0.90)
- Smoothed Geometry for Robust Attribution (NeurIPS 2020 · similarity 0.90)
- Rethinking Robustness of Model Attributions (AAAI 2024 · similarity 0.89)
- Pixel-level Certified Explanations via Randomized Smoothing (ICML 2025 · similarity 0.88)
- Benchmarking the Attribution Quality of Vision Models (NeurIPS 2024 · similarity 0.88)
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.