Research map / Adversarial robustness and security / Backdoor and poisoning
Authority Backdoor: A Certifiable Backdoor Mechanism for Authoring DNNs
Han Yang, Shaofeng Li, Tian Dong, Xiangyu Xu, Guangchi Liu, Zhen Ling
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 |
|---|---|---|---|
| add randomized smoothing (sigma=0.9) to the authority backdoor | ResNet-18 Authority Backdoor · CIFAR-10 adaptive attack | acc_reversed: 94.26% -> 14.18% | Improved |
| add randomized smoothing (sigma=0.9) to the authority backdoor | ResNet-18 Authority Backdoor · CIFAR-10 | acc_auth: 94.13% -> 78.48% | Hurt |
| add randomized smoothing | ResNet-18 Authority Backdoor · GTSRB | acc_auth: 98.55% -> 62.75% | Hurt |
| add randomized smoothing | ResNet-18 Authority Backdoor · TinyImageNet | acc_reversed: 14.10% -> 12.86% | No clear effect |
| increase smoothing noise sigma from 0.4 to 0.9 (ablation) | ResNet-18 Authority Backdoor · CIFAR-10 | acc_reversed: 54.5% -> baseline level (acc_clean) | Improved |
| increase smoothing noise sigma from 0.4 to 0.9 (ablation) | ResNet-18 Authority Backdoor · CIFAR-10 | acc_auth: 86.2% -> 73.9% | Hurt |
| add negative-loss finetuning during implanting | Authority Backdoor · clean finetuning attack, 100 samples 10 epochs | acc_clean: 0.55% -> 0.15% | Improved |
| add negative-loss finetuning during implanting | Authority Backdoor · clean finetuning attack | acc_auth: 89.33% -> 89.31% | No clear effect |
| Authority Backdoor vs AdvParams | ResNet-18 · CIFAR-10 | acc_auth: 92.02% -> 94.13% | Improved |
| Authority Backdoor vs Passports | ResNet-18 · CIFAR-10 | acc_auth: 90.89% -> 94.13% | Improved |
| Authority Backdoor vs AdvParams | ResNet-18 · GTSRB | acc_clean: 6.94% -> 9.40% | Hurt |
| apply vanilla adaptive attack to unsmoothed model | ResNet-18 Authority Backdoor · CIFAR-10 | restored accuracy: 6.02% -> 91.84% | Hurt |
Closest papers by meaning
- Untargeted Backdoor Watermark: Towards Harmless and Stealthy Dataset Copyright Protection (NeurIPS 2022 · similarity 0.89)
- Rethinking Deep Neural Network Ownership Verification: Embedding Passports to Defeat Ambiguity Attacks (NeurIPS 2019 · similarity 0.89)
- Safe and Robust Watermark Injection with a Single OoD Image (ICLR 2024 · similarity 0.87)
- EntropyMark: Towards More Harmless Backdoor Watermark via Entropy-based Constraint for Open-source Dataset Copyright Protection (CVPR 2025 · similarity 0.86)
- From Toxic to Trustworthy: Using Self-Distillation and Semi-supervised Methods to Refine Neural Networks (AAAI 2024 · similarity 0.85)
- IPRemover: A Generative Model Inversion Attack against Deep Neural Network Fingerprinting and Watermarking (AAAI 2024 · similarity 0.85)
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.
