Research map / Adversarial robustness and security / Attacks on text and graphs
ComDefend: An Efficient Image Compression Model to Defend Adversarial Examples
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 |
|---|---|---|---|
| Apply ComDefend before classification at test time | ResNet50 · CIFAR-10, FGSM strongest attack | accuracy: 35% to 83% | Helped |
| Apply ComDefend before classification at test time | ResNet50 · CIFAR-10, BIM strongest attack | accuracy: 0% to 31% | Helped |
| Apply ComDefend before classification at test time | ResNet50 · CIFAR-10, DeepFool strongest attack | accuracy: 1% to 89% | Helped |
| Apply ComDefend before classification at test time | ResNet50 · CIFAR-10, C&W strongest attack | accuracy: 0% to 87% | Helped |
| Use ComDefend instead of PixelDefend | ResNet50 · Fashion-MNIST, BIM | accuracy: 85%/83% vs 70%/60% | Hurt |
| Use ComDefend instead of HGD | IncResV2 · ImageNet, IFGSM (3/5) | accuracy: 51%/42% vs 42%/42% | Mixed |
| Use ComDefend instead of HGD | IncResV2 · ImageNet, DeepFool | accuracy: 48%/48% -> 60%/60% | Helped |
| Use 12 compressed bits rather than 8 or 10 (ablation) | ComDefend · CIFAR-10 validation images | PSNR: 31.01 -> 31.78 | Helped |
| Use 14 compressed bits rather than 12 (ablation) | ComDefend · CIFAR-10 validation images | PSNR: 31.78 -> 28.77 | Hurt |
| Use 16 compressed bits rather than 12 (ablation) | ComDefend · CIFAR-10 validation images | PSNR: 31.78 -> 30.95 | Hurt |
| Add Gaussian noise to the compact representation (ablation) | ComDefend · Image reconstruction and adversarial defense | reconstructed image quality | Helped |
| Set λ=0.01 rather than the selected λ=0.0001 (ablation) | ComDefend · CIFAR-10 training images | average accuracy: 90.60% -> 84.66% | Hurt |
Closest papers by meaning
- Feature Distillation: DNN-Oriented JPEG Compression Against Adversarial Examples (CVPR 2019 · similarity 0.89)
- PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples (ICLR 2018 · similarity 0.88)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.87)
- Countering Adversarial Images using Input Transformations (ICLR 2018 · similarity 0.87)
- Adversarial Defense by Stratified Convolutional Sparse Coding (CVPR 2019 · similarity 0.86)
- Deep Defense: Training DNNs with Improved Adversarial Robustness (NeurIPS 2018 · 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.