Research map / Adversarial robustness and security / Attacks on text and graphs
Identification of the Adversary from a Single Adversarial Example
Minhao Cheng, Rui Min, Haochen Sun, Pin-Yu Chen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Random mask watermarking embedded in head layers of each copy | model copies (VGG16, ResNet18) · CIFAR-10, GTSRB, Tiny-ImageNet | classification accuracy: worst drop about 1% | No clear effect |
| Adding data-limited adversary identification score (perturbation on masked pixels + adversarial stability) | data-limited tracing · CIFAR-10, GTSRB, Tiny-ImageNet, single adversarial example | tracing accuracy: average 71.33%, random guess 1% | Improved |
| Data-free max label score for tracing | data-free tracing · CIFAR-10, GTSRB, Tiny-ImageNet, single adversarial example | tracing accuracy: average 69.80% vs 71.33% data-limited; SignOPT up about 23 points | Mixed |
| Data-free detector instead of data-limited detector on SignOPT attack | SignOPT adversarial examples · CIFAR-10, GTSRB, Tiny-ImageNet | tracing accuracy: about 23% improvement | Improved |
| Data-limited detector applied to Bandit attack | Bandit adversarial examples · CIFAR-10, GTSRB, Tiny-ImageNet | tracing accuracy: worst case just over 42% | Hurt |
| Combining multiple adversarial examples (summing scores) (ablation) | data-limited and data-free tracing · CIFAR-10, 100 copies | tracing accuracy: around 97% with 1 example, 100% with 3 or more | Improved |
| Increasing noise level on watermark pixels (adaptive attack) on data-limited detector (ablation) | data-limited tracing · CIFAR-10 ResNet18, adaptive attack | tracing accuracy: significant accuracy drop | Hurt |
| Random-noise adaptive attack on data-free detector (ablation) | data-free tracing · CIFAR-10 ResNet18, adaptive attack | tracing accuracy: not sensitive to random noise | No clear effect |
| Increasing alpha weight on adversarial stability term (ablation) | data-limited tracing · CIFAR-10 and GTSRB, Bandit and SignOPT | tracing accuracy: qualitative | Improved |
| Increasing alpha weight on adversarial stability term (ablation) | data-limited tracing · CIFAR-10 and GTSRB, HSJ NES SimBA | tracing accuracy: slight drop | Hurt |
Closest papers by meaning
- Tracing the Origin of Adversarial Attack for Forensic Investigation and Deterrence (ICCV 2023 · similarity 0.93)
- Deep Neural Network Fingerprinting by Conferrable Adversarial Examples (ICLR 2021 · similarity 0.88)
- Rethinking Adversarial Robustness in the Context of the Right to be Forgotten (ICML 2024 · similarity 0.86)
- Detection and Defense of Unlearnable Examples (AAAI 2024 · similarity 0.85)
- Watermark-embedded Adversarial Examples for Copyright Protection against Diffusion Models (CVPR 2024 · similarity 0.85)
- The Good, the Bad and the Ugly: Meta-Analysis of Watermarks, Transferable Attacks and Adversarial Defenses (NeurIPS 2025 · similarity 0.84)
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.
