Research map / Adversarial robustness and security / Certified and empirical robustness
Retrieval-Augmented Convolutional Neural Networks Against 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 |
|---|---|---|---|
| Add retrieval-augmented classification with local mixup | Vanilla CNN · CIFAR-10, S1 direct attacks | accuracy under attack | Helped |
| Add retrieval-augmented classification with local mixup | Vanilla CNN · SVHN, S1 direct attacks | accuracy under attack | Helped |
| Add retrieval-augmented classification with local mixup | Vanilla CNN · ImageNet, S1 direct attacks | Accuracy@5 | Helped |
| Add retrieval and local mixup | Vanilla CNN · CIFAR-10, PGD | S1 / Gray-box Acc: 3.1 -> 25.9 | Helped |
| Add retrieval and local mixup | Vanilla CNN · CIFAR-10, PGD | Clean Acc: 69.5 -> 60.2 | Hurt |
| Add local mixup (ablation) | RaCNN with K=5 · CIFAR-10, S2 clean | accuracy: 72.57 -> 75.6 | Helped |
| Add local mixup (ablation) | RaCNN with K=10 · CIFAR-10, S2 clean | accuracy: 79.52 -> 80.80 | Helped |
| Increase retrieved examples from K=5 to K=10 (ablation) | RaCNN with local mixup · CIFAR-10, S2 clean | accuracy: 75.6 -> 80.80 | Helped |
| Add local mixup (ablation) | RaCNN with K=5 · SVHN, S2 clean | accuracy: 90.78 -> 91.64 | Helped |
| Add local mixup (ablation) | RaCNN with K=10 · SVHN, S2 clean | accuracy: 92.19 -> 92.49 | Helped |
| Add randomized retrieval at inference (ablation) | RaCNN · CIFAR-10, PGD | S1 / Gray-box Acc: 25.9 -> 28.4 | Helped |
| Add randomized retrieval at inference (ablation) | RaCNN · CIFAR-10, PGD | Clean Acc: 60.2 -> 60.1 | No clear effect |
Closest papers by meaning
- Boundary-Aware Backward-Compatible Representation via Adversarial Learning in Image Retrieval (CVPR 2023 · similarity 0.86)
- Defense Against Adversarial Images Using Web-Scale Nearest-Neighbor Search (CVPR 2019 · similarity 0.85)
- RetrievalGuard: Provably Robust 1-Nearest Neighbor Image Retrieval (ICML 2022 · similarity 0.84)
- Collapse-Aware Triplet Decoupling for Adversarially Robust Image Retrieval (ICML 2024 · similarity 0.83)
- Perturbation-Invariant Adversarial Training for Neural Ranking Models: Improving the Effectiveness-Robustness Trade-Off (AAAI 2024 · similarity 0.83)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.83)
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.