Research map / Adversarial robustness and security / Attacks on text and graphs
Effective Ambiguity Attack Against Passport-Based DNN Intellectual Property Protection Schemes Through Fully Connected Layer Substitution
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace passport layers with CERB blocks | Passport-protected AlexNet and ResNet18 · CIFAR10, varying passport-layer counts | ACC (%): 64.82 -> 84.22 | Helped |
| Use CERB instead of IERB | Passport-protected AlexNet and ResNet18 · CIFAR10, varying passport-layer counts | BDR: 27.95 -> 80.30 | Helped |
| Replace direct affine-factor optimization with CERB | Passport-protected ResNet18 · CIFAR100, 1000 examples (2%) | ACC (%): 53.91 -> 70.12 | Helped |
| Replace passport layers with CERB blocks | Passport-protected ResNet18 · Caltech256, last 10 passport layers | ACC (%): 24.93 -> 47.85 | Helped |
| Use CERB rather than IERB | Passport-protected ResNet18 · CIFAR10, non-overlapping data, 2000 examples (5%) | BDR: 49.20 -> 93.44 | Helped |
| Add CERB blocks at passport-layer locations (ablation) | Passport-protected ResNet18 · Caltech-256 | ACC (%): 25.03 -> 47.07 | Helped |
| Increase CERB depth beyond 2 MLP layers (ablation) | CERB attack on passport-protected ResNet18 · CIFAR10, 10 passport layers | ACC (%): 92.24 -> 86.85; 92.24 -> 84.18; 92.24 -> 86.45 | Hurt |
| Replace LeakyReLU with tanh or Sigmoid (ablation) | CERB attack on passport-protected ResNet18 · CIFAR10, 10 passport layers | ACC (%): 92.87 -> 92.54; 92.87 -> 92.80 | No clear effect |
| Use CERB with a forged signature | Greedy-Residual watermark model · Caltech256-ResNet18 | ACC (%): 54.98 -> 54.37 | No clear effect |
Closest papers by meaning
- Steganographic Passport: An Owner and User Verifiable Credential for Deep Model IP Protection Without Retraining (CVPR 2024 · similarity 0.88)
- IPRemover: A Generative Model Inversion Attack against Deep Neural Network Fingerprinting and Watermarking (AAAI 2024 · similarity 0.88)
- Safe and Robust Watermark Injection with a Single OoD Image (ICLR 2024 · similarity 0.84)
- EncryIP: A Practical Encryption-Based Framework for Model Intellectual Property Protection (AAAI 2024 · similarity 0.84)
- MAP: MAsk-Pruning for Source-Free Model Intellectual Property Protection (CVPR 2024 · similarity 0.83)
- Train to Defend: First Defense Against Cryptanalytic Neural Network Parameter Extraction Attacks (NeurIPS 2025 · similarity 0.82)
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.