AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Improving Fast Adversarial Training With Prior-Guided Knowledge

Xiaojun Jia, Yong Zhang, Xingxing Wei, Baoyuan Wu, Ke Ma, Jue Wang, Xiaochun Cao

TPAMI 2024 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements · 66 citations (OpenAlex)

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 changedWhereMeasuredEffect
add prior-guided initialization (ablation)FGSM-RS · CIFAR-10 ResNet18AutoAttack accuracy: 0.00 -> 44.92Improved
add prior-guided regularizer (ablation)FGSM-PGK initialization-only · CIFAR-10 ResNet18AutoAttack accuracy: 44.92 -> 49.02Improved
add prior-guided regularizer (ablation)FGSM-PGK initialization-only · CIFAR-10 ResNet18training time (h): 0.9 -> 1.2Hurt
add prior-guided regularizer (ablation)FGSM-PGK initialization-only · CIFAR-10 ResNet18clean accuracy: 84.92 -> 81.77Hurt
add prior-guided ensemble weights (ablation)FGSM-PGK initialization-only · CIFAR-10 ResNet18AutoAttack accuracy: 44.92 -> 46.08Improved
add prior-guided weights to initialization plus regularizer (ablation)FGSM-PGK · CIFAR-10 ResNet18AutoAttack accuracy: 49.02 -> 49.51Improved
replace fixed-decay EMA with dynamic decay rate EMA (D-EMA) (ablation)FGSM-RS-EMA · CIFAR-10 ResNet18AutoAttack accuracy (best): 45.15 -> 46.63Improved
replace fixed-decay EMA with D-EMA (ablation)FGSM-RS-EMA · CIFAR-10 ResNet18AutoAttack accuracy (last): 0.0 -> 46.57Improved
use random initialization instead of zero initializationFGSM-AT · CIFAR-10catastrophic overfitting epoch: 20-th epoch -> 70-th epochMixed
use FGSM-PGK instead of multi-step PGD-ATResNet18 · CIFAR-10AutoAttack accuracy: 48.93 -> 49.51Improved
use FGSM-PGK instead of multi-step PGD-ATResNet18 · CIFAR-10training time (h): 4.4 -> 1.2Improved
use FGSM-PGK instead of FGSM-PGIWideResNet34-10 · CIFAR-10PGD-50 accuracy: 56.4 -> 59.79Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy