AI Paper Atlas

Research map / Adversarial robustness and security / Black-box and transfer attacks

Deep Neural Network Fingerprinting by Conferrable Adversarial Examples

Nils Lukas, Yuxuan Zhang, Florian Kerschbaum

ICLR 2021 · Adversarial robustness and security › Black-box and transfer attacks · approach: Transfer attacks on face recognition · 22 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 CEM-generated conferrable examples as the fingerprint instead of IPGuard examplesCIFAR-10 surrogate verification · CIFAR-10 retrained modelsROC AUC: 0.63 -> 1.0Improved
Generate examples with CEM at ε=0.01 rather than ε=0.15CEM · CIFAR-10mean conferrability: 0.24 -> 0.49Improved
Use a larger perturbation threshold, from ε=0.01 to ε=0.15CEM fingerprint · CIFAR-10 surrogate modelsCAEAcc: 0.85 -> 0.98Improved
Use a larger perturbation threshold, from ε=0.01 to ε=0.15CEM fingerprint · CIFAR-10 reference modelsCAEAcc: 0.42 -> 0.75Hurt
Detect fingerprint queries using the evaluated detectorCEM fingerprint verification · CIFAR-10, ε≤0.025ROC AUC: 0.67No clear effect
Apply fine-tuning and pruning to the source-derived modelCEM fingerprint · CIFAR-10fingerprint removabilityNo clear effect
Extract surrogates through retraining and related extraction attacksCEM fingerprint · CIFAR-10fingerprint removabilityNo clear effect
Transfer learn from ImageNet32 using CIFAR-10 domain dataCEM fingerprint · CIFAR-10fingerprint removabilityHurt
Transfer learn without CIFAR-10 data, using CINIC dataCEM fingerprint · CIFAR-10fingerprint removabilityNo clear effect
Adversarially train the surrogate from scratchCEM fingerprint · CIFAR-10, ε=0.025CAEAcc: 15%Hurt
Give the extractor more ground-truth labelsCEM fingerprint · CIFAR-10 surrogate modelsCAEAccHurt
Set α=1.0 and β=γ=0.1 in the CEM objective (ablation)CEM · CIFAR-10, ε=0.025mean conferrability: 0.40 -> 0.49Improved

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