Research map / Adversarial robustness and security / Certified and empirical robustness
Consistency Regularization for 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 |
|---|---|---|---|
| Add adversarial consistency regularization across two independent augmentations | Standard AT · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 44.86 -> 56.38 | Helped |
| Add adversarial consistency regularization across two independent augmentations | Standard AT · CIFAR-100, PreAct-ResNet-18 | AutoAttack robust accuracy: 19.85 -> 24.55 | Helped |
| Add adversarial consistency regularization | TRADES · CIFAR-10, l1 PGD-100 at epsilon 4000/255 | robust accuracy: 29.58 -> 48.32 | Helped |
| Add consistency regularization | Standard AT · CIFAR-10-C | mCE: 24.03 -> 21.83 | Helped |
| Add Cutout augmentation (ablation) | Standard AT · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 44.86 -> 49.95 | Helped |
| Add AutoAugment (ablation) | Standard AT · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 44.86 -> 55.18 | Helped |
| Add consistency regularization after AutoAugment (ablation) | Standard AT · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 55.18 -> 56.38 | Helped |
| Use temperature 0.5 rather than 0.8 (ablation) | Consistency-regularized Standard AT · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 56.22 -> 56.38 | Helped |
| Replace JS discrepancy with MSE (ablation) | TRADES + consistency · CIFAR-10, PreAct-ResNet-18 | AutoAttack robust accuracy: 47.68 -> 47.30 | Hurt |
| Replace JS discrepancy with KL divergence (ablation) | TRADES + consistency · CIFAR-10, PreAct-ResNet-18 | AutoAttack robust accuracy: 47.68 -> 43.25 | Hurt |
| Match clean augmented predictions rather than independently attacked predictions (ablation) | Prior consistency regularization · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 55.18 -> 53.11 | Hurt |
| Apply the consistency loss to a single augmented instance (ablation) | Standard AT · CIFAR-10, PreAct-ResNet-18 | PGD-100 robust accuracy: 53.20 | Hurt |
Closest papers by meaning
- Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples (ICML 2023 · similarity 0.93)
- Data Augmentation Can Improve Robustness (NeurIPS 2021 · similarity 0.92)
- Towards Better Robust Generalization with Shift Consistency Regularization (ICML 2021 · similarity 0.91)
- Towards Understanding the Regularization of Adversarial Robustness on Neural Networks (ICML 2020 · similarity 0.91)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.91)
- Efficient local linearity regularization to overcome catastrophic overfitting (ICLR 2024 · similarity 0.90)
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.