AI Paper Atlas

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

Can't Steal? Cont-Steal! Contrastive Stealing Attacks Against Image Encoders

Zeyang Sha, Xinlei He, Ning Yu, Michael Backes, Yang Zhang

CVPR 2023 · Adversarial robustness and security › Attacks on text and graphs · approach: Neural network watermarking · 24 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 contrastive stealing rather than conventional embedding-matchingMoCo ResNet18 encoder stealing · CIFAR10 downstream; STL10 surrogateagreement: 0.479 -> 0.841Improved
Use contrastive stealing rather than conventional embedding-matchingSimCLR ResNet18 encoder stealing · STL10 downstream; STL10 surrogateaccuracy: 0.457 -> 0.714Improved
Use contrastive stealing rather than INFONCEImage encoder stealing · CIFAR10 downstream; CIFAR10 surrogateagreement: 0.826 -> 0.845Improved
Use contrastive stealing rather than INFONCEImage encoder stealing · STL10 downstream; STL10 surrogateagreement: 0.806 -> 0.829Improved
Replace training from scratch with Cont-StealSimCLR ResNet18 encoder · CIFAR10 pretrainingtraining time (h): 20.01 -> 0.62Improved
Add Gaussian noise with standard deviation 10 to target embeddingsCont-Steal · STL10 surrogate and downstream; SimCLR pretrained on CIFAR10accuracy: 0.729 -> 0.410Hurt
Add negative pairs generated by the surrogate encoder (ablation)Cont-Steal with BYOL · STL10 downstream; STL10 surrogateagreement: 0.314 -> 0.817Improved
Add target-encoder negative pairs (ablation)Cont-Steal with BYOL · CIFAR10 surrogateagreement: 0.242 -> 0.844Improved
Add target-encoder negative pairs (ablation)Cont-Steal with BYOL · STL10 surrogateagreement: 0.314 -> 0.817Improved
Use a ResNet50 surrogate instead of ResNet18 (ablation)Cont-Steal on SimCLR · CIFAR10 surrogateaccuracy: 0.839 -> 0.840No clear effect
Use DenseNet161 rather than ResNet18 as the surrogate (ablation)Cont-Steal on SimCLR · CIFAR10 surrogateaccuracy: 0.839 -> 0.828Hurt
Use MobileNetV2 rather than ResNet18 as the surrogate (ablation)Cont-Steal on SimCLR · CIFAR10 surrogateaccuracy: 0.839 -> 0.811Hurt

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