Research map / Federated learning and privacy / Personalized and heterogeneous FL
FLSeg: Enhancing Privacy and Robustness in Federated Learning under Heterogeneous Data via Model Segmentation
Zichun Su, Zhi Lu, Yutong Wu, Renfei Shen, Songfeng Lu
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 |
|---|---|---|---|
| apply Krum on model segments (Segment Aggregation) instead of whole model | Krum · CIFAR-100 Min-Max 20% | top-1 accuracy (%): 16.38 -> 46.05 | Improved |
| apply Krum on model segments instead of whole model | Krum · CIFAR-10 Min-Max 20% | top-1 accuracy (%): 45.46 -> 60.82 | Improved |
| apply DnC on model segments instead of whole model | DnC · CIFAR-10 Min-Max 20% | top-1 accuracy (%): 50.62 -> 64.59 | Improved |
| apply DnC on model segments instead of whole model | DnC · CIFAR-100 Min-Max 20% | top-1 accuracy (%): 34.23 -> 51.11 | Improved |
| use DnC (SVD spectral) instead of Krum inside FLSeg | FLSeg · CIFAR-10 Min-Max 20% | top-1 accuracy (%): 60.82 -> 64.59 | Improved |
| replace ACORN with FLSeg(DnC) under Min-Max attackers | ACORN · CIFAR-10 Min-Max 20% | top-1 accuracy (%): 48.21 -> 64.59 | Improved |
| replace PEFL with FLSeg(DnC) under Min-Max attackers | PEFL · FEMNIST Min-Max 20% | top-1 accuracy (%): 21.32 -> 77.74 | Improved |
| replace ACORN with FLSeg(DnC) with no attack | ACORN · CIFAR-10 no attack | top-1 accuracy (%): 49.08 -> 63.99 | Improved |
| use Segment Exchange with PRNG seeds instead of SMPC (ACORN) for privacy, client side | ACORN · n=1000, l=1e6 | client computation time (ms): 1696 ms -> 233 ms | Improved |
| use Segment Exchange instead of SMPC for server-side privacy cost | ACORN · n=1000, l=1e6 | server time per client (ms): 181 ms -> 78 ms | Improved |
| choose segment size 10^4 instead of whole model (ablation) | FLSeg(DnC) · CIFAR-10 Min-Max | top-1 accuracy (%): 57.75% -> 65.72% | Improved |
| reduce segment size from 10^4 to 10^2 (ablation) | FLSeg(DnC) · CIFAR-10 Min-Max | top-1 accuracy (%): 65.72% -> 63.93% | Hurt |
Closest papers by meaning
- FedLMT: Tackling System Heterogeneity of Federated Learning via Low-Rank Model Training with Theoretical Guarantees (ICML 2024 · similarity 0.89)
- A Bayesian Approach for Personalized Federated Learning in Heterogeneous Settings (NeurIPS 2024 · similarity 0.88)
- FedP3: Federated Personalized and Privacy-friendly Network Pruning under Model Heterogeneity (ICLR 2024 · similarity 0.88)
- FedAA: A Reinforcement Learning Perspective on Adaptive Aggregation for Fair and Robust Federated Learning (AAAI 2025 · similarity 0.87)
- Federated Learning via Input-Output Collaborative Distillation (AAAI 2024 · similarity 0.87)
- Towards Trustworthy Federated Learning with Untrusted Participants (ICML 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.
