Research map / Federated learning and privacy / Privacy attacks and secure learning
FedCut: A Spectral Analysis Framework for Reliable Detection of Byzantine Colluders
Hanlin Gu, Lixin Fan, Xingxing Tang, Qiang Yang
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 spatial-temporal consistency across iterations (FedCut vs per-iteration NCut) (ablation) | FedCut · MNIST/Fashion-MNIST Non-IID, 30 attackers, Fang-v1 | model performance: 31.7% (NCut) vs FedCut stable | Improved |
| use FedCut instead of robust-statistics defenses under Fang-v1 attack | Median · MNIST IID, 30 attackers | model performance: 43.4 -> 92.3 | Improved |
| use FedCut instead of Bulyan under Fang-v2 attack | Bulyan · MNIST IID, 30 attackers | model performance: 18.8 -> 92.3 | Improved |
| use FedCut instead of DnC under Gaussian attack | DnC · MNIST IID, 30 attackers | model performance: 74.3 -> 92.3 | Improved |
| use FedCut instead of Kmeans under multi-group collusion | Kmeans · Fashion-MNIST/MNIST collusion | model performance: 22.4 -> 91.9 worst-case MNIST | Improved |
| increase number of attackers from 10 to 30 | Median · Fashion-MNIST | averaged model performance: 88.6% -> 74.6% | Hurt |
| increase number of attackers | FedCut · Fashion-MNIST | averaged model performance | No clear effect |
| make client data more heterogeneous (IID to beta=0.1) | Trimmed Mean · Fashion-MNIST | averaged model performance: 74.9% -> 55.4% | Hurt |
| make client data more heterogeneous | FedCut · Fashion-MNIST | model performance: drops no more than 3% | No clear effect |
| increase number of colluder groups from 2 to 3 | DnC · MNIST Non-IID beta=0.5, 30 attackers | model performance: 91.9 -> 87.2 | Hurt |
| vary number of colluder groups from 2 to 5 | FedCut · MNIST Non-IID beta=0.5 | model performance: 91.7, 91.6, 92.1, 91.7 | No clear effect |
| use FedCut spectral heuristics on toy mimic collusion scenario S3 | FedCut vs Median · toy example 1000 runs | Byzantine tolerant rate: 37.7% -> 98.7% | Improved |
Closest papers by meaning
- Efficient Federated Learning against Byzantine Attacks and Data Heterogeneity via Aggregating Normalized Gradients (NeurIPS 2025 · similarity 0.86)
- Rethinking Byzantine Robustness in Federated Recommendation from Sparse Aggregation Perspective (AAAI 2025 · similarity 0.85)
- Dynamic Byzantine-Robust Learning: Adapting to Switching Byzantine Workers (ICML 2024 · similarity 0.85)
- Byzantine-Robust Federated Learning: Impact of Client Subsampling and Local Updates (ICML 2024 · similarity 0.85)
- Byzantine-Robust Decentralized Learning via Remove-then-Clip Aggregation (AAAI 2024 · similarity 0.84)
- Exploit Gradient Skewness to Circumvent Byzantine Defenses for Federated Learning (AAAI 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.
