Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Efficient and Effective Adversarial Training
Gaurang Sriramanan, Sravanti Addepalli, Arya Baburaj, Venkatesh Babu Radhakrishnan
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 exponential weight averaging (ablation) | NuAT · CIFAR-10 WRN-34-10 AutoAttack v2 | robust accuracy (%): 50.81 -> 53.36 | Improved |
| add exponential weight averaging (ablation) | NuAT · CIFAR-10 ResNet-18 AutoAttack v1 | robust accuracy (%): 49.24 -> 50.75 | Improved |
| add a second attack step (cross-entropy only) (ablation) | NuAT · CIFAR-10 WRN-34-10 AutoAttack v2 | robust accuracy (%): 50.81 -> 51.27 | Improved |
| add weight-averaged model supervision to two-step attack (ablation) | NuAT2 · CIFAR-10 WRN-34-10 AutoAttack v2 | robust accuracy (%): 51.27 -> 54.76 | Improved |
| add hybrid adaptive switching to two-step (ablation) | NuAT · CIFAR-10 WRN-34-10 AutoAttack v2 | robust accuracy (%): 50.81 -> 51.58 | Improved |
| replace GAT l2 regularizer with Nuclear-Norm regularizer | single-step AT · CIFAR-10 ResNet-18 AutoAttack v1 | robust accuracy (%): 47.30 -> 49.24 | Improved |
| replace GAT l2 regularizer with Nuclear-Norm regularizer | single-step AT · ImageNet-100 AutoAttack | robust accuracy (%): 28.92 -> 31.96 | Improved |
| replace GAT l2 regularizer with Nuclear-Norm regularizer | single-step AT · MNIST AutoAttack | robust accuracy (%): 90.62 -> 93.11 | Improved |
| use Nuclear Norm instead of Frobenius norm in attack objective | attack generation · CIFAR-10 normal model | average class entropy: 0.87 -> 0.97 | Improved |
| use single-step NuAT instead of 10-step PGD-AT | adversarial training · CIFAR-10 ResNet-18 AutoAttack v1 | robust accuracy (%): 48.75 -> 49.24 | Improved |
| use single-step NuAT-WA instead of AWP | adversarial training · CIFAR-10 WRN-34-10 AutoAttack v2 | robust accuracy (%): 56.17 (AWP) vs 53.36 (NuAT-WA) | Hurt |
Closest papers by meaning
- Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples (ICML 2023 · similarity 0.91)
- Efficient Robust Training via Backward Smoothing (AAAI 2022 · similarity 0.89)
- Variational Adversarial Defense: A Bayes Perspective for Adversarial Training (TPAMI 2023 · similarity 0.89)
- Adversarial Robustness through Local Linearization (NeurIPS 2019 · similarity 0.89)
- Clarifying the Behavior and the Difficulty of Adversarial Training (AAAI 2024 · similarity 0.88)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 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.
