Research map / Federated learning and privacy / Privacy attacks and secure learning
Sibai: A Few-Shot Meta-Classifier for Poisoning Detection in Federated Learning
Melanie Götz, Torsten Krauß, Alexandra Dmitrienko
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 |
|---|---|---|---|
| Sibai instead of MESAS | FL poisoning detection · Scenario 4 (0.9 1-class non-i.i.d.), random label flip | F1: 0.0000 -> 0.8235 | Improved |
| Sibai instead of MESAS | FL poisoning detection · Scenario 1 (i.i.d. low PDR), blend backdoor | F1: 0.1818 -> 1.0000 | Improved |
| Sibai instead of MESAS | FL poisoning detection · Scenario 6 (inter-client non-i.i.d.), semantic backdoor | F1: 0.3157 -> 0.9473 | Improved |
| Sibai instead of HDBSCAN | FL poisoning detection · Scenario 11 (adaptive), pixel backdoor | F1: 1.0000 -> 0.9411 | Hurt |
| Sibai instead of MESAS | FL poisoning detection · Scenario 11 (adaptive), pixel backdoor | F1: 0.6666 -> 0.9411 | Improved |
| lower poison data rate (high to low PDR) vs Sibai | Sibai · i.i.d. CIFAR-10, semantic backdoor | F1: 1.0000 -> 0.8750 | Hurt |
| increase 1-class non-i.i.d. degree q from 0.6 to 0.9 | Sibai · random label flip, high PDR | F1: 1.0000 -> 0.8235 | Hurt |
| add cosine and Euclidean distance features to PCA input (ablation) | Sibai · detection accuracy | detection accuracy: up to 10% | Improved |
| attacker adapts to cosine distance (alpha 0.3) | Sibai · i.i.d. scenarios 8 and 9 | F1: F1 1.0000 for all attacks | No clear effect |
| attacker adapts to Euclidean distance (alpha 0.3), low PDR | Sibai · scenario 11, pixel and blend | F1: F1 1.0000 -> 0.9411 | Mixed |
| use Sibai instead of FLTrust | FL defense · Scenario 4 pixel backdoor | backdoor accuracy (BA, lower better): BA 53.1 -> 0.0 | Improved |
| use Sibai instead of Krum | FL defense · Scenario 1 random label flip | MA diff: MA diff -35.0 -> -0.8 | Improved |
Closest papers by meaning
- Tracing Back the Malicious Clients in Poisoning Attacks to Federated Learning (NeurIPS 2025 · similarity 0.89)
- Dual Defense: Enhancing Privacy and Mitigating Poisoning Attacks in Federated Learning (NeurIPS 2024 · similarity 0.88)
- Model Poisoning Attacks to Federated Learning via Multi-Round Consistency (CVPR 2025 · similarity 0.88)
- FedREDefense: Defending against Model Poisoning Attacks for Federated Learning using Model Update Reconstruction Error (ICML 2024 · similarity 0.87)
- Backdoor Federated Learning by Poisoning Backdoor-Critical Layers (ICLR 2024 · similarity 0.87)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (NeurIPS 2025 · similarity 0.87)
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.
