Research map / Adversarial robustness and security / Certified and empirical robustness
Masking Adversarial Damage: Finding Adversarial Saliency for Robust and Sparse Network
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 |
|---|---|---|---|
| Use MAD saliency to prune parameters at 90% sparsity rather than keeping a full model | Adversarially trained networks · CIFAR-10 and SVHN, three architectures | Adversarial robustness: 4.2(%) and 1.9(%) degradation for CIFAR-10 and SVHN each on average of three networks | Mixed |
| Use MAD pruning instead of LWM, ADMM, or HYDRA | Adversarially trained networks · CIFAR-10 and SVHN | Adversarial robustness degradation: MAD has 4.2(%) and 1.9(%) degradation; HYDRA has 7.2(%) and 6.1(%) | Helped |
| Prune parameters in ascending rather than descending adversarial saliency order (ablation) | MAD · CIFAR-10, VGG-16, 90% sparsity | PGD accuracy: 43.1 -> 51.8 | Helped |
| Prune parameters in ascending rather than descending adversarial saliency order (ablation) | MAD · CIFAR-10, VGG-16, 99% sparsity | PGD accuracy: 33.9 -> 41.0 | Helped |
| Use low-saliency pruning instead of random pruning (ablation) | MAD · CIFAR-10, VGG-16, 90% sparsity | PGD accuracy: 43.5 -> 51.8 | Helped |
| Use low-saliency pruning instead of random pruning (ablation) | MAD · CIFAR-10, VGG-16, 99% sparsity | PGD accuracy: 31.3 -> 41.0 | Helped |
Closest papers by meaning
- Adversarial Neural Pruning with Latent Vulnerability Suppression (ICML 2020 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Sparse DNNs with Improved Adversarial Robustness (NeurIPS 2018 · similarity 0.89)
- Nasty Adversarial Training: A Probability Sparsity Perspective for Robustness Enhancement (ICLR 2026 · similarity 0.89)
- A Combination of Noise and Bilateral Filters Achieve Supralinear and Scalable Adversarial Robustness in CNNs (CVPR 2026 · similarity 0.88)
- On the Vulnerability of Adversarially Trained Models Against Two-faced Attacks (ICLR 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.