Research map / Adversarial robustness and security / Certified and empirical robustness
Explaining Classifiers Using Adversarial Perturbations on the Perceptual Ball
Check 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 perceptual-loss regularization to adversarial perturbations | ImageNet weak localization · ImageNet val, first 2000 images | best-threshold error | Helped |
| Add gradient guidance to perceptual perturbations (ablation) | ImageNet weak localization · ImageNet val, first 2000 images | value-threshold error: 0.44 -> 0.41 | Helped |
| Add perceptual loss rather than using unregularized perturbations (ablation) | Perceptual perturbation explanation · ImageNet val, 5000 images | deletion AUC: 0.10 -> 0.07 | Helped |
| Add Gaussian blur to the saliency map (ablation) | Perceptual perturbation explanation · ImageNet val, 5000 images | insertion AUC: 0.54 -> 0.63 | Helped |
| Add Gaussian blur to the saliency map (ablation) | Perceptual perturbation explanation · ImageNet val, 5000 images | deletion AUC: 0.07 -> 0.12 | Hurt |
| Add perceptual loss rather than using unregularized perturbations (ablation) | VOC pointing game · VOC pointing game, original image | standard success rate: 81.2 -> 85.1 | Helped |
| Add perceptual loss rather than using unregularized perturbations (ablation) | VOC pointing game · VOC pointing game, original image | difficult-subset success rate: 62.9 -> 69.0 | Helped |
| Resize input images to 1.5x before computing the pointing-game saliency map | Pointing-game methods · VOC pointing game | standard success rate: 85.1 -> 88.2 | Helped |
| Regularize more and higher network layers (ablation) | Perceptual perturbations · Weak localization, insertion/deletion, and pointing game | object localization | Helped |
| Use only the top layer for perceptual regularization (ablation) | Perceptual perturbations · Pointing game | success rate | Helped |
| Restrict blur width to σ ∈ {0,…,25} (ablation) | Perceptual perturbations · VOC pointing game, original image | standard success rate: 0.3% | Hurt |
| Restrict blur width to σ ∈ {0,…,25} (ablation) | Perceptual perturbations · VOC pointing game, original image, difficult subset | success rate | Helped |
Closest papers by meaning
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.88)
- Theoretical Understanding of Learning from Adversarial Perturbations (ICLR 2024 · similarity 0.87)
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.87)
- Towards Fine-Grained Interpretability: Counterfactual Explanations for Misclassification with Saliency Partition (CVPR 2025 · similarity 0.86)
- Rethinking Visual Counterfactual Explanations Through Region Constraint (ICLR 2025 · similarity 0.86)
- Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective (NeurIPS 2024 · 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.