AI Paper Atlas

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

Robustness Tokens: Towards Adversarial Robustness of Transformers

Brian Pulfer, Yury Belousov, Slava Voloshynovskiy

ECCV 2024 · Adversarial robustness and security › Certified and empirical robustness · approach: Robustness-accuracy trade-off analysis · 4 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
append 10 learned Robustness Tokens to DiNOv2 inputDiNOv2 (S/B/L/G) · ImageNet-1k linear classification (clean)accuracy: average -0.5Hurt
append 10 learned Robustness Tokens to DiNOv2 inputDiNOv2 (S/B/L/G) · ADE20K linear segmentation (clean)mIoU: +0.2No clear effect
add Robustness TokensDiNOv2-B · ImageNet PGD attack classificationaccuracy: 0.0 -> 50.0Improved
add Robustness TokensDiNOv2-B · ADE20K PGD attack segmentationmIoU: 3.1 -> 23.4Improved
add Robustness Tokens to larger modelsDiNOv2-S vs G · ImageNet PGD attack classificationaccuracy: 31.9 (S) vs 63.1 (G)Improved
increase number of Robustness Tokens beyond 10 (1, 10, 20, 50) (ablation)DiNOv2 · training losstraining lossNo clear effect
use a single Robustness Token (ablation)DiNOv2 · feature orientationtraining lossImproved
tokens trained on PGD, tested with AutoAttackDiNOv2-L · ImageNet val AutoAttackaccuracy: 1.8 -> 60.1Improved
add Robustness Tokens to OpenCLIPOpenCLIP Base · clean-vs-adversarial cosine similarityavg cosine similarity: -0.02 -> 0.86Improved
add Robustness Tokens to DEIT-IIIDEIT-III Base · clean-vs-adversarial cosine similarityavg cosine similarity: 0.16 -> 0.74Improved
train tokens for about 200 steps with batch size 8Robustness Tokens · ImageNet subset (1600 images)training convergence: converges in under 200 stepsImproved

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