AI Paper Atlas

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

NeurIPS 2021 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements · 30 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 exponential weight averaging (ablation)NuAT · CIFAR-10 WRN-34-10 AutoAttack v2robust accuracy (%): 50.81 -> 53.36Improved
add exponential weight averaging (ablation)NuAT · CIFAR-10 ResNet-18 AutoAttack v1robust accuracy (%): 49.24 -> 50.75Improved
add a second attack step (cross-entropy only) (ablation)NuAT · CIFAR-10 WRN-34-10 AutoAttack v2robust accuracy (%): 50.81 -> 51.27Improved
add weight-averaged model supervision to two-step attack (ablation)NuAT2 · CIFAR-10 WRN-34-10 AutoAttack v2robust accuracy (%): 51.27 -> 54.76Improved
add hybrid adaptive switching to two-step (ablation)NuAT · CIFAR-10 WRN-34-10 AutoAttack v2robust accuracy (%): 50.81 -> 51.58Improved
replace GAT l2 regularizer with Nuclear-Norm regularizersingle-step AT · CIFAR-10 ResNet-18 AutoAttack v1robust accuracy (%): 47.30 -> 49.24Improved
replace GAT l2 regularizer with Nuclear-Norm regularizersingle-step AT · ImageNet-100 AutoAttackrobust accuracy (%): 28.92 -> 31.96Improved
replace GAT l2 regularizer with Nuclear-Norm regularizersingle-step AT · MNIST AutoAttackrobust accuracy (%): 90.62 -> 93.11Improved
use Nuclear Norm instead of Frobenius norm in attack objectiveattack generation · CIFAR-10 normal modelaverage class entropy: 0.87 -> 0.97Improved
use single-step NuAT instead of 10-step PGD-ATadversarial training · CIFAR-10 ResNet-18 AutoAttack v1robust accuracy (%): 48.75 -> 49.24Improved
use single-step NuAT-WA instead of AWPadversarial training · CIFAR-10 WRN-34-10 AutoAttack v2robust accuracy (%): 56.17 (AWP) vs 53.36 (NuAT-WA)Hurt

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