AI Paper Atlas

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

Adversarial Purification by Consistency-Aware Latent Space Optimization on Data Manifolds

Shuhai Zhang, Jiahao Yang, Hui Luo, Jie Chen, Ke Wang, Feng Liu, Bo Han, Mingkui Tan

TPAMI 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Augmentation and diffusion purification · 0 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
replace consistency model with GAN latent space (ablation)CMAP · CIFAR-10 PGD+EOTrobust accuracy (%): 74.60 -> 40.80Hurt
replace consistency model with VAE (ablation)CMAP · CIFAR-10 PGD+EOTrobust accuracy (%): 74.60 -> 58.80Hurt
replace consistency model with VAE (ablation)CMAP · ImageNet-100 PGD+EOTrobust accuracy (%): 39.87 -> 43.60; standard 61.93 -> 68.20Mixed
add DiffPure on top of CMAPCMAP · CIFAR-10 BPDA+EOTrobust accuracy (%): 83.53 -> 89.47Improved
add DiffPure on top of CMAPCMAP · CIFAR-10 standard accuracyaccuracy (%): 88.73 -> 91.33Improved
add DiffPure on top of CMAPCMAP · CIFAR-10 PGD-l2+EOTrobust accuracy (%): 84.00 -> 83.60No clear effect
use CMAP instead of best purification baselineCMAP · CIFAR-10 PGD+EOTrobust accuracy (%): 55.87 -> 74.60Improved
use CMAP instead of FreqPureCMAP · ImageNet-100 BPDA+EOTrobust accuracy (%): 19.58 -> 49.47Improved
use CMAP instead of GNSPCMAP · CIFAR-10 BPDA+EOTrobust accuracy (%): 88.53 -> 83.53Hurt
use CMAP instead of GNSP at larger epsilon 16/255CMAP · CIFAR-10 AutoAttackrobust accuracy (%): 33.80 -> 80.73Improved
use CMAP instead of FreqPureCMAP · CIFAR-10 AutoAttack-l2robust accuracy (%): 88.26 -> 83.40Hurt
use CMAP instead of FreqPureCMAP · CIFAR-10 clean accuracyaccuracy (%): 93.96 -> 88.73Hurt

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