Research map / Federated learning and privacy / Privacy attacks and secure learning
Mind the Cost of Scaffold! Benign Clients May Even Become Accomplices of Backdoor Attack
Xingshuo Han, Xuanye Zhang, Xiang Lan, Haozhao Wang, Shengmin Xu, Shen Ren, Jason Zeng, Ming Wu, Michael Heinrich, Tianwei Zhang
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 GAN-based data supplementation to the attacker dataset (ablation) | BadSFL · CIFAR-10 | primary task accuracy: below 25% -> around 55% | Improved |
| add GAN-based data supplementation to the attacker dataset (ablation) | BadSFL · MNIST | primary task accuracy: under 75% -> above 90% | Improved |
| tamper with control variate and optimize for predicted future global model | BadSFL vs Black-box and Neurotoxin · CIFAR-10/CIFAR-100 | backdoor durability: below 50% after the 60th round -> exceeding 90% over 100 rounds | Improved |
| use feature-based trigger instead of label flipping or pattern trigger (ablation) | BadSFL · CIFAR-10 | backdoor task accuracy | Improved |
| BadSFL versus baselines under defenses (DP, pruning, FLAME, SparseFed) | BadSFL · CIFAR-10/CIFAR-100 | attack success rate | Improved |
| use Neurotoxin's parameter selection for durability | Neurotoxin · SFL CIFAR | backdoor accuracy after attacker exit | No clear effect |
| BadSFL versus baselines on MNIST pattern trigger | BadSFL · MNIST | backdoor accuracy: 10% higher accuracy | Improved |
| BadSFL versus baselines on MNIST label flipping after attacker exit | BadSFL · MNIST | backdoor persistence: 5 times longer-lasting | Improved |
Closest papers by meaning
- Eliminate Distance Differences Induced by Backdoor Attacks: Layer-Selective Training and Clipping to Mask Backdoor Models (CVPR 2026 · similarity 0.87)
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 2024 · similarity 0.85)
- Backdoor Federated Learning by Poisoning Backdoor-Critical Layers (ICLR 2024 · similarity 0.85)
- Infighting in the Dark: Multi-Label Backdoor Attack in Federated Learning (CVPR 2025 · similarity 0.84)
- Protecting against simultaneous data poisoning attacks (ICLR 2025 · similarity 0.84)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (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.
