Research map / Adversarial robustness and security / Certified and empirical robustness
Evaluating the Robustness of Neural Networks: An Extreme Value Theory Approach
Read the paperCheck 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 defensive distillation to the CNN | MNIST CNN · MNIST | CLEVER ℓ2: 0.721 -> 0.865 | Helped |
| Add defensive distillation to the CNN | CIFAR CNN · CIFAR-10 | CLEVER ℓ2: 0.072 -> 0.130 | Helped |
| Replace ReLU with bounded ReLU | MNIST CNN · MNIST | CLEVER ℓ2: 0.721 -> 0.833 | Helped |
| Replace ReLU with bounded ReLU | CIFAR CNN · CIFAR-10 | CLEVER ℓ2: 0.072 -> 0.045 | Hurt |
| Estimate the local cross-Lipschitz maximum using reverse Weibull extreme-value fitting | CLEVER · MNIST, CIFAR-10, ImageNet | Reverse Weibull goodness-of-fit acceptance: Table percentages are mostly near 100.0 | Helped |
| Reduce the number of batches used to estimate CLEVER | CLEVER on most models · MNIST and CIFAR-10 | CLEVER score stability | No clear effect |
Closest papers by meaning
- Measuring Neural Net Robustness with Constraints (NeurIPS 2016 · similarity 0.91)
- Accurate, reliable and fast robustness evaluation (NeurIPS 2019 · similarity 0.91)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.90)
- Benchmarking Adversarial Robustness on Image Classification (CVPR 2020 · similarity 0.90)
- A Recipe for Improved Certifiable Robustness (ICLR 2024 · similarity 0.90)
- On Explaining Neural Network Robustness with Activation Path (ICLR 2023 · similarity 0.90)
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.