Research map / Adversarial robustness and security / Certified and empirical robustness
Disentangling Adversarial Robustness and Generalization
Check 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 |
|---|---|---|---|
| Increase training data for a fixed model | Classifiers on FONTS, EMNIST, and Fashion-MNIST · FONTS, EMNIST, Fashion-MNIST | regular adversarial attack success rate | No clear effect |
| Increase training data | Classifiers evaluated with on-manifold attacks · FONTS, EMNIST, Fashion-MNIST | on-manifold adversarial attack success rate | Helped |
| Add on-manifold adversarial training using the true manifold | Classifier · FONTS | test error | Helped |
| Add on-manifold adversarial training using an approximated manifold | Classifier · EMNIST | test error | Helped |
| Add adversarial transformation training using known invariances | Classifier · EMNIST, Fashion-MNIST | test error | Helped |
| Use random data augmentation instead of adversarial transformations (ablation) | Classifier · EMNIST, Fashion-MNIST | test error | Hurt |
| Add regular adversarial training | Classifier · FONTS | test accuracy: roughly twice the amount of training data | Hurt |
| Combine regular and on-manifold adversarial training | Classifier · FONTS | regular robustness and generalization | Mixed |
| Use class-agnostic rather than class-specific VAE-GANs for on-manifold training (ablation) | On-manifold adversarial training · FONTS, EMNIST | test error | Hurt |
| Use randomly perturbed examples instead of adversarially perturbed examples (ablation) | Regular and on-manifold adversarial training · FONTS, EMNIST | generalization | Hurt |
| Use the weak regular adversarial-training variant, stopping at the first adversarial example (ablation) | Regular adversarial training · FONTS, EMNIST | generalization | Helped |
| Use the full or weak variant of on-manifold adversarial training (ablation) | On-manifold adversarial training · FONTS, EMNIST | test error | No clear effect |
Closest papers by meaning
- Fantastic Robustness Measures: The Secrets of Robust Generalization (NeurIPS 2023 · similarity 0.92)
- Dual Manifold Adversarial Robustness: Defense against Lp and non-Lp Adversarial Attacks (NeurIPS 2020 · similarity 0.91)
- Explaining the role of Intrinsic Dimensionality in Adversarial Training (ICML 2025 · similarity 0.90)
- Adversarially Robust Generalization Requires More Data (NeurIPS 2018 · similarity 0.90)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · similarity 0.89)
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · 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.