AI Paper Atlas

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

Defense Against Adversarial Attacks via Controlling Gradient Leaking on Embedded Manifolds

Yueru Li, Shuyu Cheng, Hang Su, Jun Zhu

ECCV 2020 · Adversarial robustness and security › Certified and empirical robustness · approach: Attack evaluation and defense critique · 18 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 PCA projection to 800 dims with small-scale noise (noise800) over ordinary trainingDenseNet · CIFAR10 C&W l2mean perturbation norm: 0.090 -> 0.208Improved
add PCA projection to 300 dims with noise (noise300) over ordinary trainingDenseNet · CIFAR10 cleanerror rate (%): 5.92 -> 16.53Hurt
add noise in small-scale PCA directions on top of PCA projection (800 dims) (ablation)DenseNet · CIFAR10 C&W l2mean perturbation norm: 0.160 -> 0.208Improved
add noise in small-scale PCA directions on top of PCA projection (300 dims) (ablation)DenseNet · CIFAR10 C&W l2mean perturbation norm: 0.256 -> 0.308Improved
add noise in small-scale directions on top of PCA projection (800 dims) (ablation)DenseNet · CIFAR10 alpha300gradient leaking alpha300: 0.244 -> 0.460Improved
reduce PCA dimension from 800 to 300 (no noise) (ablation)DenseNet · CIFAR10 cleanerror rate (%): 6.97 -> 11.71Hurt
reduce PCA dimension from 800 to 300 (no noise) (ablation)DenseNet · CIFAR10 C&W l2mean perturbation norm: 0.160 -> 0.256Improved
use noise800 instead of TRADES adversarial trainingDenseNet · CIFAR10 cleanerror rate (%): 21.22 -> 8.82Improved
use noise800 instead of TRADES adversarial trainingDenseNet · CIFAR10 C&W l2mean perturbation norm: 0.664 -> 0.208Hurt
add PCA-projection preprocessing to MMC loss (MMC-P-500)MMC · CIFAR10 untargeted PGD10 eps=8robust accuracy (%): 25.6 -> 42.5Improved
add PCA-projection preprocessing to MMC loss (MMC-P-300)MMC · CIFAR10 untargeted PGD50 eps=8robust accuracy (%): 9.2 -> 29.2Improved
add PCA-projection preprocessing to MMC loss (MMC-P-300)MMC · CIFAR10 cleanaccuracy (%): 92.5 -> 87.9Hurt

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