Research map / Adversarial robustness and security / Certified and empirical robustness
On the Minimal Adversarial Perturbation for Deep Neural Networks With Provable Estimation Error
Fabio Brau, Giulio Rossolini, Alessandro Biondi, Giorgio Buttazzo
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 |
|---|---|---|---|
| use CB strategy instead of FOB (Bisection) | boundary distance estimation · MNIST | average distance from boundary: 1.803954 -> 1.645090 | Improved |
| use CB strategy instead of FOB (Bisection) | boundary distance estimation · GTSRB | average distance from boundary: 4.885275 -> 3.665061 | Improved |
| use Bisection CB instead of Newton CB for sigma-hat | boundary estimation · FMNIST (rho=sqrt2) | sigma-hat: 0.02 -> 0.06 | Improved |
| use CB instead of FOB for sigma-hat | boundary estimation · MNIST (rho=2) | sigma-hat: 0.37 -> 0.72 | Improved |
| use CB instead of FOB for sigma-hat | boundary estimation · GTSRB (rho=2) | sigma-hat: 0.49 -> 0.87 | Improved |
| use DeepFool instead of CB (global vs local method) | distance estimation · Fashion-MNIST | average distance from boundary: 0.739743 -> 0.548780 | Mixed |
| use Newton instead of Bisection algorithm for CB | zero finding · FMNIST (rho=sqrt2) | sigma-hat: 0.06 -> 0.02 | Hurt |
| test attacks bounded by t(x)/rho* below sigma-hat | CB Bisection · MNIST and GTSRB | attack success | Improved |
| test attacks bounded by t(x)/rho* below sigma-hat | CB Bisection · FMNIST and CIFAR10 | attack success | Mixed |
Closest papers by meaning
- Lower bounds on the robustness to adversarial perturbations (NeurIPS 2017 · similarity 0.91)
- Adversarial Robustness Guarantees for Random Deep Neural Networks (ICML 2021 · similarity 0.91)
- Scaling provable adversarial defenses (NeurIPS 2018 · similarity 0.89)
- How Many Perturbations Break This Model? Evaluating Robustness Beyond Adversarial Accuracy (ICML 2023 · similarity 0.89)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · 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.
