Research map / Adversarial robustness and security / Certified and empirical robustness
Generalizable Adversarial Training via Spectral Normalization
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 spectral normalization (ablation) | AlexNet with FGM ℓ2 adversarial training · CIFAR10 | test accuracy: 0.54 -> 0.63 | Helped |
| Add spectral normalization (ablation) | AlexNet with PGM ℓ2 adversarial training · CIFAR10 | test accuracy: 0.50 -> 0.62 | Helped |
| Add spectral normalization (ablation) | AlexNet with WRM adversarial training · CIFAR10 | test accuracy: 0.61 -> 0.65 | Helped |
| Add spectral normalization (ablation) | AlexNet with ERM · CIFAR10 | test accuracy: 0.79 -> 0.79 | No clear effect |
| Add spectral normalization (ablation) | 1-layer MLP with FGM ℓ2 adversarial training · MNIST | test accuracy: 0.88 -> 0.96 | Helped |
| Add spectral normalization (ablation) | AlexNet with FGM ℓ2 adversarial training · SVHN | test accuracy: 0.76 -> 0.83 | Helped |
| Normalize the full convolution operation rather than only the convolution kernel (ablation) | AlexNet with PGM ℓ2 adversarial training · CIFAR10 | validation accuracy: 0.55 -> 0.60 | Helped |
| Use spectral normalization rather than batch normalization (ablation) | AlexNet with PGM ℓ2 adversarial training · CIFAR10 | validation accuracy: 0.54 -> 0.60 | Helped |
| Use spectral normalization rather than weight decay (ablation) | AlexNet with PGM ℓ2 adversarial training · CIFAR10 | validation accuracy: 0.55 -> 0.60 | Helped |
| Use spectral normalization rather than dropout (ablation) | AlexNet with PGM ℓ2 adversarial training · CIFAR10 | validation accuracy: 0.52 -> 0.60 | Helped |
| Apply spectral normalization to adversarial training (ablation) | AlexNet with FGM, PGM, and WRM training · CIFAR10 | test accuracy: 0.09, 0.11, and 0.04 | Helped |
Closest papers by meaning
- Intriguing Properties of Adversarial Training at Scale (ICLR 2020 · similarity 0.90)
- Spectral Normalization for Generative Adversarial Networks (ICLR 2018 · similarity 0.90)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.88)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.87)
- Boosting Adversarial Training via Fisher-Rao Norm-based Regularization (CVPR 2024 · similarity 0.87)
- Dynamical Low-Rank Compression of Neural Networks with Robustness under Adversarial Attacks (NeurIPS 2025 · similarity 0.87)
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.