Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Adversarial Robustness by Enforcing Local and Global Compactness
Anh Bui, Trung Le, He Zhao, Paul Montague, Olivier deVel, Tamas Abraham, Dinh Phung
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 local compactness (ablation) | ADV · MNIST (PGD-40, eps=0.325) | robust accuracy: 88.1% -> 91.43% | Improved |
| add global compactness (ablation) | ADR+LC · MNIST (PGD-40, eps=0.325) | robust accuracy: 91.43% -> 94.52% | Improved |
| add clustering assumption (ablation) | ADR+LC/GB · MNIST (PGD-40, eps=0.325) | robust accuracy: 94.52% -> 94.96% | Improved |
| add compactness components to ADV (ablation) | ADV · MNIST (PGD-40, eps=0.325) | natural accuracy: 99.27% -> 99.36% | No clear effect |
| add compactness and clustering terms to ADV | ADV · MNIST (eps=0.3, eta=0.01) | robust accuracy: 94.15% -> 96.55% | Improved |
| add compactness and clustering terms to TRADES | TRADES · MNIST (eps=0.3, eta=0.01) | robust accuracy: 93.64% -> 95.71% | Improved |
| add compactness terms to ADV | ADV · CIFAR-10 (standard CNN) | multi-targeted robust accuracy: 18.73% -> 22.58% | Improved |
| add compactness terms to TRADES | TRADES · CIFAR-10 (standard CNN) | multi-targeted robust accuracy: 18.01% -> 26.7% | Improved |
| add compactness terms to TRADES | TRADES · CIFAR-10 (standard CNN) | natural accuracy: 71.37% -> 69.0% | Hurt |
| add compactness terms to ADV | ADV · CIFAR-10 (ResNet) | non-targeted robust accuracy: 44.08% -> 44.72% | No clear effect |
| add compactness terms to TRADES | TRADES · CIFAR-10 (ResNet) | non-targeted robust accuracy: 37.52% -> 40.17% | Improved |
| add compactness terms to ADV | ADV · CIFAR-10 (ResNet) | natural accuracy: 78.84% -> 78.43% | No clear effect |
Closest papers by meaning
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.91)
- Explaining Adversarial Robustness of Neural Networks from Clustering Effect Perspective (ICCV 2023 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
- Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples (ICML 2023 · similarity 0.90)
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.89)
- An Introduction to Adversarially Robust Deep Learning (TPAMI 2023 · similarity 0.89)
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.
