AI Paper Atlas

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

Catastrophic Child's Play: Easy to Perform, Hard to Defend Adversarial Attacks

Chih-Hui Ho, Brandon Leung, Erik Sandstrom, Yen Chang, Nuno Vasconcelos

CVPR 2019 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial perturbations beyond images · 7 citations (OpenAlex)

Check 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 pose variation rather than camera shake perturbationsImageNet-trained CNNs · ImageNetrecognition rate: 73.7 -> 47.2Hurt
Add frontal-pose real-world examples to defense trainingCNNs under camera shake attacks · ImageNet to Frontalrecognition rate: 73.7 -> 82.0Improved
Add frontal-pose real-world examples to defense trainingCNNs under pose variation attacks · ImageNet to Frontalrecognition rate: 47.2 -> 63.7Improved
Use all camera-shake and pose-variation defense images instead of frontal-only imagesCNNs under camera shake attacks · Frontal to Allrecognition rate: 82.0 -> 87.1Improved
Use all camera-shake and pose-variation defense images instead of frontal-only imagesCNNs under pose variation attacks · Frontal to Allrecognition rate: 63.7 -> 79.1Improved
Apply Color Jitter defenseCNNs under camera shake attacks · Avgrecognition rate: 80.9 -> 82.7No clear effect
Use gradient-based defenses instead of no defenseCNNs under camera shake attacks · Avgrecognition rate: 80.9 -> 80.0Hurt
Use gradient-based defenses instead of no defenseCNNs under pose variation attacks · Avgrecognition rate: 63.3 -> 59.3Hurt
Craft attacks by selecting the most adversarial instance from K=5 candidates rather than sampling randomlyCNNs under camera shake attacks · ImageNetrecognition rate: 73.7 -> 51.3Hurt
Craft attacks by selecting the most adversarial instance from K=5 candidates rather than sampling randomlyCNNs under pose variation attacks · ImageNetrecognition rate: 47.2 -> 33.0Hurt
Use camera shake rather than pose variation as an object-recognition perturbationTurk semantic perceptibility assessment · Object recognition Turk experimentSIPR (%): 81.5 -> 91.6Improved

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