Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Robustness of Deep Neural Networks via Regularization
Yao Li, Martin Renqiang Min, Thomas Lee, Wenchao Yu, Erik Kruus, Wei Wang, Cho-Jui Hsieh
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 embedding regularization (ER-Classifier vs Madry's Adv) | Madry's Adv · CIFAR10 AutoAttack | robust accuracy (%): 35.6 -> 47.2 | Improved |
| add embedding regularization (ER-Classifier vs Madry's Adv) | Madry's Adv · MNIST AutoAttack | robust accuracy (%): 69.0 -> 79.0 | Improved |
| add embedding regularization (ER-Classifier vs Madry's Adv) | Madry's Adv · Tiny ImageNet AutoAttack | robust accuracy (%): 9.70 -> 11.1 | Improved |
| use ER-Classifier instead of Defense-GAN | Defense-GAN · MNIST C&W | accuracy (%): 55.0 -> 99.1 | Improved |
| add discriminator-based embedding regularization to encoder-classifier (E-CLA) (ablation) | E-CLA · MNIST, CIFAR10, Tiny ImageNet PGD | robust accuracy | Improved |
| add discriminator-based embedding regularization to encoder-classifier (E-CLA) (ablation) | E-CLA · STL10 PGD | robust accuracy | Improved |
| use Wasserstein discriminator regularization instead of VAE KL regularization (ablation) | VAE-CLA · four datasets PGD | robust accuracy | Improved |
| train without min-max robust optimization (ER-Classifier-) on CIFAR10 under strong attack (ablation) | ER-Classifier · CIFAR10 PGD | robust accuracy | Mixed |
| add ER-Detector combined code (ER2) versus direct output (ER1) vs KD in familiar PGD detection | KD · MNIST PGD | ACC-COM (%): 29.87 -> 63.81 | Improved |
| use ER-Detector instead of KD when trained on PGD, attacked by C&W | KD · MNIST cross-attack | ACC-COM (%): 10.05 -> 100.00 | Improved |
| use ER-Detector instead of KD when trained on PGD, attacked by FGSM | KD · CIFAR10 cross-attack | ACC-NADV (%): 97.89 -> 98.19 | Mixed |
Closest papers by meaning
- Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples (ICML 2023 · similarity 0.89)
- Towards Understanding the Regularization of Adversarial Robustness on Neural Networks (ICML 2020 · similarity 0.89)
- On Connections Between Regularizations for Improving DNN Robustness (TPAMI 2020 · similarity 0.89)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.87)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · similarity 0.86)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.86)
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.
