AI Paper Atlas

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

Self-Supervised Representation Learning for Adversarial Attack Detection

Yi Li, Plamen Angelov, Neeraj Suri

ECCV 2024 · Adversarial robustness and security › Attacks on text and graphs · approach: Defenses against universal perturbations · 12 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 supervised and pre-trained detectors with self-supervised PAA-ResNet-Ladversarial attack detection · ImageNet-Rdetection rate on attacked images (%): 82.8 -> 86.8 (vs DINOV2)Improved
PAA-ResNet-S vs DINOV2 on costdetector · ImageNet-Rlatency (ms): 572.0 -> 6.2Improved
add prototype-wise contrastive estimation L_PCE on top of L_PM (ablation)PAA-ResNet-L · ImageNet-Rattacked detection rate (%): 68.9 -> 82.8Improved
add instance-wise contrastive loss L_ICL with discrimination bank (ablation)L_PM + L_PCE · ImageNet-Rattacked detection rate (%): 82.8 -> 86.8Improved
replace L_PM with L_ICL in combination with L_PCE (ablation)L_PCE + L_ICL · ImageNet-Rattacked detection rate (%): 82.8 -> 82.0Hurt
use L_PCE instead of prototype loss L_S2Z (ablation)L_PM-based training · ImageNet-Rattacked detection rate (%): 68.6 -> 72.4Improved
add data augmentation alone to baseline (ablation)PAA-ResNet-L · ImageNet-Rattacked detection rate (%): 52.3 -> 50.1Hurt
use PAA blocks instead of axial attention (ablation)ResNet-50, HRNet, Xception · ImageNet-Rdetection accuracy: 2.1% to 2.9% over axial-attentionImproved
vary temperature tau from 0.03 to 0.2L_PM · ImageNet-Rdetection accuracyNo clear effect
weight L_PCE and L_ICL tenfold relative to L_PMtotal loss · ImageNet-Rdetection accuracyNo clear effect
test on unseen dataset instead of seen datasetall detectors · cross-dataset (Table 6)detection rateHurt

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