Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Examples Are a Natural Consequence of Test Error in Noise
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 |
|---|---|---|---|
| train with Gaussian noise augmentation (σ=0.4) | InceptionV3 · ImageNet-C in-memory, all corruptions | accuracy: 45.0 -> 52.6 | Helped |
| train with Gaussian noise augmentation (σ=0.1) | Wide-ResNet-28-10 · CIFAR-10-C, all corruptions | accuracy: 76.3 -> 81.2 | Helped |
| adversarial training | Wide-ResNet-28-10 · CIFAR-10-C, Blur | accuracy: 85.5 -> 80.2 | Mixed |
| train with Gaussian noise augmentation (σ=0.1) | Wide-ResNet-28-10 · CIFAR-10-C, Speckle | accuracy: 62.8 -> 91.1 | Helped |
| adversarial training | Wide-ResNet-28-10 · CIFAR-10-C, Contrast | accuracy: 92.2 -> 77.0 | Hurt |
| increase Gaussian augmentation scale from σ=0.1 to σ=0.4 (ablation) | Wide-ResNet-28-10 · CIFAR-10-C, all corruptions | accuracy: 81.2 -> 74.7 | Hurt |
| train with Gaussian noise augmentation (σ=0.4) | Wide-ResNet-28-10 · CIFAR-10-C, Contrast | accuracy: 92.2 -> 41.5 | Hurt |
| adversarial training | MNIST classifier · MNIST, Gaussian noise σ=0.3 | accuracy: 96.9% -> 78.2% | Hurt |
| train with Gaussian noise augmentation (σ=0.4) | InceptionV3 · ImageNet-C compressed, all corruptions | accuracy: 38.8 -> 42.7 | Helped |
Closest papers by meaning
- The Odds are Odd: A Statistical Test for Detecting Adversarial Examples (ICML 2019 · similarity 0.90)
- Adversarial Examples Are Not Bugs, They Are Features (NeurIPS 2019 · similarity 0.89)
- Adversarial Risk and the Dangers of Evaluating Against Weak Attacks (ICML 2018 · similarity 0.88)
- Detecting Adversarial Examples Is (Nearly) As Hard As Classifying Them (ICML 2022 · similarity 0.88)
- Are adversarial examples inevitable? (ICLR 2019 · similarity 0.87)
- Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective (NeurIPS 2024 · similarity 0.85)
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.