Research map / Adversarial robustness and security / Certified and empirical robustness
What Can the Neural Tangent Kernel Tell Us About Adversarial Robustness?
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 |
|---|---|---|---|
| Generate adversarial perturbations from the NTK instead of using the trained network's own attack gradients | Wide two-layer neural networks in the kernel regime · CIFAR-10 car vs plane | robust accuracy: eventually driving robust accuracy to 0% | Hurt |
| Use only robust NTK-induced features and discard non-robust features (ablation) | NTK feature-based classifiers · MNIST and CIFAR-10 | classification performance | Mixed |
| Replace the cross-entropy-derived NTK perturbation with the alternative Taylor-expansion perturbation (ablation) | NTK adversarial attacks · Kernel-regime attack experiments | attack effect | No clear effect |
| Use adversarial training rather than standard training | Deep convolutional network empirical NTK · CIFAR-10 | concentration on top eigenvalues | Helped |
| Use adversarial training rather than standard training | Deep convolutional network empirical NTK · CIFAR-10 and MNIST | time to enter the lazy regime | Helped |
| Train longer with standard training, allowing later-learned non-robust features to contribute | Standardly trained neural networks · MNIST and CIFAR-10 | robustness | Hurt |
Closest papers by meaning
- Theoretical Analysis of Robust Overfitting for Wide DNNs: An NTK Approach (ICLR 2024 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.88)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.88)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.88)
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.