AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

Diffusion Models for Adversarial Purification

Weili Nie, Brandon Guo, Yujia Huang, Chaowei Xiao, Arash Vahdat, Animashree Anandkumar

ICML 2022 · Adversarial robustness and security › Attacks on text and graphs · approach: Diffusion-based adversarial purification · 76 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
Use DiffPure instead of the strongest baseline without extra dataWideResNet-28-10 · CIFAR-10, AutoAttack ℓ∞ ε=8/255robust accuracy: 65.24 -> 70.64±0.39Improved
Use DiffPure instead of the strongest baseline without extra dataWideResNet-70-16 · CIFAR-10, AutoAttack ℓ∞ ε=8/255robust accuracy: 66.60 -> 71.29±0.55Improved
Use DiffPure instead of the strongest baseline without extra dataWideResNet-28-10 · CIFAR-10, AutoAttack ℓ2 ε=0.5robust accuracy: 78.32 -> 78.58±0.40Improved
Use DiffPure instead of the strongest baseline without extra dataWideResNet-70-16 · CIFAR-10, AutoAttack ℓ2 ε=0.5robust accuracy: 80.86 -> 80.60±0.57Mixed
Use DiffPure instead of the strongest baselineResNet-50 · ImageNet, AutoAttack ℓ∞ ε=4/255robust accuracy: 37.89 -> 40.93±1.96Improved
Use DiffPure instead of the baselineWideResNet-50-2 · ImageNet, AutoAttack ℓ∞ ε=4/255robust accuracy: 39.25 -> 44.39±0.95Improved
Use DiffPure instead of the baselineDeiT-S · ImageNet, AutoAttack ℓ∞ ε=4/255robust accuracy: 35.50 -> 43.18±1.27Improved
Use DiffPure instead of the strongest unseen-threat baselineResNet-50 · CIFAR-10, AutoAttack ℓ∞ ε=8/255robust accuracy: 40.0 -> 70.0±1.2Improved
Use DiffPure instead of the strongest unseen-threat baselineResNet-50 · CIFAR-10, AutoAttack ℓ2 ε=1robust accuracy: 34.9 -> 70.9±0.6Improved
Use DiffPure instead of the strongest purification baselineEyeglasses attribute classifier · CelebA-HQ, BPDA+EOT ℓ∞ ε=16/255robust accuracy: 75.00 -> 90.63±1.10Improved
Use DiffPure instead of DSM+LDWideResNet-28-10 · CIFAR-10, BPDA+EOT ℓ∞ ε=8/255robust accuracy: 70.01 -> 81.40±0.16Improved
Replace VP-SDE sampling with VP-ODE sampling (ablation)DiffPure · CIFAR-10, AutoAttack ℓ∞ ε=8/255robust accuracy: 70.64±0.39 -> 39.86±0.98Hurt

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