AI Paper Atlas

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

Reliably Fast Adversarial Training via Latent Adversarial Perturbation

Geon Yeong Park, Sang Wan Lee

ICCV 2021 · Adversarial robustness and security › Certified and empirical robustness · approach: Regularized defenses against attacks · 23 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 latent adversarial perturbation (SLAT) to single-step FGSM AT (ablation)FGSM AT · CIFAR-10PGD-50-10 robust accuracy (%): 0.01 -> 47.06Improved
add latent adversarial perturbation to FGSM-RS (ablation)FGSM-RS · CIFAR-10PGD-50-10 robust accuracy (%): 3.90 -> 44.95Improved
derive latent perturbation from the original sample rather than the randomly perturbed one (SLAT vs FGSM-RS + latent) (ablation)SLAT · CIFAR-10PGD-50-10 robust accuracy (%): 44.95 -> 47.06Improved
reduce FGSM step size to 0.9*8/255 instead of using latent perturbation (ablation)FGSM AT · CIFAR-10PGD-50-10 robust accuracy (%): 0.01 -> 3.09No clear effect
inject latent perturbation into conv3, conv4 instead of conv1, conv2 (ablation)SLAT · CIFAR-10PGD-50-10 robust accuracy (%): 47.06 -> 0.00Hurt
inject latent perturbation into all of conv1-conv4 in addition to input (ablation)SLAT · CIFAR-10standard accuracy (%): 85.91 -> 81.41Hurt
inject latent perturbation into all of conv1-conv4 in addition to input (ablation)SLAT · CIFAR-10AutoAttack robust accuracy (%): 44.62 -> 43.58Hurt
inject latent perturbation into all of conv1-conv4 in addition to input (ablation)SLAT · CIFAR-10PGD-50-10 robust accuracy (%): 47.06 -> 47.3No clear effect
replace SLAT with FGSM-CKPTSLAT vs FGSM-CKPT · CIFAR-10PGD-50-10 robust accuracy (%): 40.83 -> 47.06Improved
compare SLAT with Free-AT (m=8)SLAT vs Free-AT · CIFAR-10PGD-50-10 robust accuracy (%): 44.15 -> 47.06Improved
compare SLAT with PGD-7 ATSLAT vs PGD-7 · CIFAR-10training time (min): 383.2 -> 104.6Improved
compare SLAT with FGSM-GASLAT vs FGSM-GA · CIFAR-10training time (min): 297.9 -> 104.6Improved

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