Research map / Adversarial robustness and security / Attacks on text and graphs
SafetyNet: Detecting and Rejecting Adversarial Examples Robustly
Jiajun Lu, Theerasit Issaranon, David Forsyth
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 classification-confidence rejection (ablation) | SafetyNet m-SVM detector · CIFAR-10 DeepFool Type II | % misclassified and undetected: 22.01 -> 8.32 | Improved |
| add classification-confidence rejection (ablation) | SafetyNet m-SVM detector · CIFAR-10 L0 Type II | % misclassified and undetected: 19.73 -> 9.86 | Improved |
| add classification-confidence rejection (ablation) | SafetyNet m-SVM detector · ImageNet-1000 DeepFool Type II | % misclassified and undetected: 45.26 -> 2.32 | Improved |
| add classification-confidence rejection (ablation) | SafetyNet m-SVM detector · ImageNet-1000 L2 Type II | % misclassified and undetected: 26.40 -> 15.05 | Improved |
| add classification-confidence rejection (ablation) | Subnet detector · CIFAR-10 L0 Type II | % misclassified and undetected: 20.73 -> 19.69 | Improved |
| use quantized RBF-SVM detector instead of detection subnetwork | detector · CIFAR-10 DeepFool Type II | % misclassified and undetected: 96.24 -> 22.01 | Improved |
| use quantized RBF-SVM detector instead of detection subnetwork | detector · ImageNet-1000 L0 Type II | % misclassified and undetected: 70.52 -> 25.15 | Improved |
| use quantized RBF-SVM detector instead of detection subnetwork | detector · CIFAR-10 L2 Type II | % misclassified and undetected: 12.30 -> 18.70 | Hurt |
| use quantized RBF-SVM detector instead of detection subnetwork | detector · CIFAR-10 Fast Type II | % misclassified and undetected: 1.89 -> 6.86 | Hurt |
| add detectors on fc6 and pool5 to fc7 detector (Det ABC vs Det A) (ablation) | SafetyNet on SceneProof · SceneProof Type I average F->T | rate: 6.0% -> 5.0% | Improved |
| add detectors on fc6 and pool5 to fc7 detector (Det ABC vs Det A) (ablation) | SafetyNet on SceneProof · SceneProof Type II average F->T | rate: 9.6% -> 8.4% | Improved |
| add SafetyNet detector A to VGG19 | VGG19 classifier · SceneProof Type I average F->T | rate: 28.6% -> 6.0% | Improved |
Closest papers by meaning
- On the Robustness of Semantic Segmentation Models to Adversarial Attacks (CVPR 2018 · similarity 0.88)
- Natural Adversarial Examples (CVPR 2021 · similarity 0.88)
- All You Need Is RAW: Defending against Adversarial Attacks with Camera Image Pipelines (ECCV 2022 · similarity 0.88)
- Exploiting Multi-Object Relationships for Detecting Adversarial Attacks in Complex Scenes (ICCV 2021 · similarity 0.88)
- Adversarial Attacks Are Reversible With Natural Supervision (ICCV 2021 · similarity 0.88)
- Harnessing Edge Information for Improved Robustness in Vision Transformers (AAAI 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.
