AI Paper Atlas

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

Enhance the Visual Representation via Discrete Adversarial Training

Xiaofeng Mao, YueFeng Chen, Ranjie Duan, Yao Zhu, Gege Qi, Shaokai Ye, Xiaodan Li, Rong Zhang, Hui Xue'

NeurIPS 2022 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial perturbations beyond images · 10 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 DAT (discrete adversarial training)ResNet50 · ImageNet FGSMaccuracy (%): 12.19 -> 30.66Improved
add DATAugReg-ViT · ImageNet valtop-1 accuracy (%): 79.91 -> 81.46Improved
add DATAugReg-ViT · ImageNet-Atop-1 accuracy (%): 19.03 -> 30.15Improved
add DATMAE-H finetuning · ImageNet valtop-1 accuracy (%): 86.90 -> 87.02Improved
add DAT in pre-trainingMoCov3 · ImageNet linear evaltop-1 accuracy (%): 68.63 -> 69.60Improved
add DATSimSiam · ADE20K mIoU / FlowersmIoU: 0.2212 -> 0.2203Mixed
add DAT instead of Det-AdvPropEfficientDet-Lite0 · COCO-CmAP: 16.11 -> 17.89Improved
remove adversarial component (alpha=0, VQGAN augmentation only) (ablation)DAT ResNet50 · ImageNet FGSMaccuracy (%): 30.66 -> 23.94Hurt
increase perturbation magnitude alpha from 0.1 to 0.4 (ablation)DAT ResNet50 · ImageNet valtop-1 accuracy (%): 76.52 -> 74.28Hurt
replace adversarial discrete perturbation with random discrete perturbation (ablation)DAT ResNet50 · ImageNet FGSMaccuracy (%): 30.66 -> 29.01Hurt
keep discretizer Q at inference (ablation)DAT ResNet50 · ImageNetclean top-1 accuracy (%) and FGSM: 76.52 -> 74.8 clean; 30.66 -> 55.4 FGSMMixed
bound delta with l_inf eps=1/255 (ablation)DAT ResNet50 · ImageNet valtop-1 accuracy (%): 76.52 -> 76.10Hurt

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