Research map / Adversarial robustness and security / Certified and empirical robustness
A Bayes-Optimal View on Adversarial Examples
Eitan Richardson, Yair Weiss
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 RBF SVM (appropriate bandwidth) in place of CNN | CNN · Symmetric CelebA/MNIST datasets | adversarial accuracy: CelebA Male: 0% -> 8% | Improved |
| add RBF SVM in place of CNN | CNN · Real CelebA Eyeglasses | adversarial accuracy: 0.00% -> 32.3% | Improved |
| add RBF SVM in place of CNN | CNN · Real CelebA Bangs | adversarial accuracy: 6.00% -> 24.7% | Improved |
| add RBF SVM in place of Linear SVM | Linear SVM · Real CelebA Smiling | adversarial accuracy: 11.7% -> 16.3% | Improved |
| train CNN on symmetric data instead of real data, test on real images | CNN · CelebA Heavy Makeup test accuracy | test accuracy: 86.3% -> 79.9% | Hurt |
| train CNN on symmetric data instead of real data, test on real images | CNN · CelebA Eyeglasses test accuracy | test accuracy: 98.4% -> 91.7% | Hurt |
| train CNN on symmetric data instead of real data, test on real images | CNN · CelebA Smiling test accuracy | test accuracy: 89.5% -> 85.6% | Hurt |
| add more training examples to CNN (up to 1 million) (ablation) | CNN · Symmetric datasets | adversarial robustness: up to 1 million examples | No clear effect |
| increase CNN filters, layers, training iterations (ablation) | CNN · Symmetric datasets | adversarial robustness | No clear effect |
| add TRADES adversarial training to CNN | CNN · Symmetric datasets | robustness and accuracy | Mixed |
| use larger RBF bandwidth (ablation) | RBF SVM · Symmetric datasets | adversarial robustness | Hurt |
Closest papers by meaning
- On the Existence of The Adversarial Bayes Classifier (NeurIPS 2021 · similarity 0.92)
- Are adversarial examples inevitable? (ICLR 2019 · similarity 0.89)
- Adversarial examples from computational constraints (ICML 2019 · similarity 0.89)
- Are Generative Classifiers More Robust to Adversarial Attacks? (ICML 2019 · similarity 0.89)
- Adversarial Examples Are Not Bugs, They Are Features (NeurIPS 2019 · 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.
