Research map / Adversarial robustness and security / Attacks on text and graphs
Consensus Learning with Multi-Party Perturbation Triggers for Secure Model Access
Yizhun Zhang, Jie Huang, Zeping Zhang, Shuaishuai Zhang, Changhao Ding, Xuan 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 |
|---|---|---|---|
| protect model with CPTM consensus triggers | ResNet18 · CIFAR-10 (5 clients) | accuracy on authorized data: 93.08 -> 93.33 | No clear effect |
| protect model with CPTM consensus triggers | VGG16 · CIFAR-100 (5 clients) | accuracy on authorized data: 70.19 -> 67.63 | Hurt |
| protect model with CPTM consensus triggers | DenseNet · CIFAR-10 (5 clients) | accuracy on authorized data: 95.39 -> 93.45 | Hurt |
| feed original unauthorized inputs to the protected model | CPTM protected models · CIFAR-10, CIFAR-100, Face-LFW | accuracy: 93.33 -> 0.56 | Hurt |
| attacker uses a single client's perturbation to craft triggers (ablation) | CPTM protected ResNet18 · CIFAR-10 | attacker accuracy: 0.56 -> 1.90 | No clear effect |
| attacker fuses perturbations from three of five clients (ablation) | CPTM protected VGG16 · CIFAR-10 | attacker accuracy: 1.31 -> 15.08 | Improved |
| attacker uses a single client's perturbation (ablation) | CPTM protected DenseNet · MNIST | attacker accuracy: 10.37 -> 17.16 | Improved |
| increase number of clients from 5 to 20 (ablation) | CPTM on ResNet18 · CIFAR-10 | Acc-Pro: 93.33 -> 93.02 | No clear effect |
| increase number of clients from 5 to 20 (ablation) | CPTM on ResNet18 · CIFAR-10 | Acc-Origin: 0.56 -> 0.72 | No clear effect |
| attacker extracts trigger features from partial consensus data and applies them to own samples | CPTM protected models · trigger extraction attack, four datasets | attacker accuracy | No clear effect |
Closest papers by meaning
- Learning Mutual View Information Graph for Adaptive Adversarial Collaborative Perception (CVPR 2026 · similarity 0.84)
- Incentives in Private Collaborative Machine Learning (NeurIPS 2023 · similarity 0.83)
- Provable Robustness against Backdoor Attacks via the Primal-Dual Perspective on Differential Privacy (NeurIPS 2026 · similarity 0.82)
- MAP: MAsk-Pruning for Source-Free Model Intellectual Property Protection (CVPR 2024 · similarity 0.82)
- SPMC: Self-Purifying Federated Backdoor Defense via Margin Contribution (ICML 2025 · similarity 0.81)
- GroupCover: A Secure, Efficient and Scalable Inference Framework for On-device Model Protection based on TEEs (ICML 2024 · similarity 0.81)
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.
