Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Defense by Stratified Convolutional Sparse Coding
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 |
|---|---|---|---|
| project inputs with clustered STL and retrain the classifier on projected images rather than use PixelDefend | VGG-16 · CIFAR-10, FGSM-0.08 | accuracy: 0.8316 -> 0.8567 | Helped |
| use cluster-specific filters instead of universal STL filters (ablation) | STL · CIFAR-10, BIM | accuracy: 0.8754 -> 0.8803 | Helped |
| use cluster-specific filters instead of universal STL filters (ablation) | STL · ImageNet-10 64, BIM | accuracy: 0.8164 -> 0.8103 | Hurt |
| use cluster-specific filters instead of universal STL filters (ablation) | STL · ImageNet-10 128, BIM | accuracy: 0.8320 -> 0.8385 | Helped |
| increase the number of filters (ablation) | STL · ImageNet-10, FGSM-0.08 | accuracy on transformed adversarial data | Hurt |
| increase the sparsity constraint weight lambda (ablation) | STL · ImageNet-10 64, FGSM-0.08 | accuracy on transformed clean data | Hurt |
| increase the sparsity constraint weight lambda (ablation) | STL · ImageNet-10 64, FGSM-0.08 | gap between clean and adversarial accuracy | Helped |
| vary filter size (ablation) | STL · ImageNet-10 64 | prediction accuracy | No clear effect |
| use clustered STL rather than MagNet under BPDA | VGG-16 · CIFAR-10, BPDA | accuracy: 0.10 -> 0.42* | Helped |
| use STL defense rather than no defense against BPDA | VGG-16 · ImageNet, BPDA | accuracy | No clear effect |
| train with FGSM examples rather than use attack-agnostic STL for the designated FGSM attack | CIFAR-10, VGG-16 · FGSM-0.08 | accuracy: 0.8567 -> 0.8890 | Helped |
| use attack-agnostic STL rather than FGSM adversarial training on an unknown universal perturbation | CIFAR-10, VGG-16 · Uni-0.08 | accuracy: 0.8282 -> 0.8642 | Helped |
Closest papers by meaning
- Barrage of Random Transforms for Adversarially Robust Defense (CVPR 2019 · similarity 0.87)
- Enhancing Transformation-Based Defenses Against Adversarial Attacks with a Distribution Classifier (ICLR 2020 · similarity 0.87)
- Adversarial Robustness via Deformable Convolution with Stochasticity (ICML 2025 · similarity 0.87)
- Transferable Structural Sparse Adversarial Attack Via Exact Group Sparsity Training (CVPR 2024 · similarity 0.86)
- Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations (ICML 2024 · similarity 0.86)
- GSE: Group-wise Sparse and Explainable Adversarial Attacks (ICLR 2025 · similarity 0.85)
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.