Research map / Federated learning and privacy / Personalized and heterogeneous FL
SHE-LoRA: Selective Homomorphic Encryption for Federated Tuning with Heterogeneous LoRA
Jianmin Liu, Li Yan, Borui Li, Lei Yu, Chao Shen
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 |
|---|---|---|---|
| use SHE-LoRA (selective encryption with column swapping) instead of MaskCrypt | federated LoRA OpenLLaMA-3B · DAGER on SST2, B=4 | ROUGE-1 reconstruction score (lower is better): 89.16 -> 0.72 | Improved |
| use SHE-LoRA instead of Flex-LoRA-DP noise | federated LoRA · DAGER on SST2, B=16 | ROUGE-1: 68.62 -> 0.0 | Improved |
| use DP noise on gradients instead of no protection | Flex-LoRA · DAGER on SST2, B=16 | ROUGE-1: 10.27 -> 68.62 | Hurt |
| use SHE-LoRA instead of MaskCrypt | federated LoRA · DAGER on Rotten Tomatoes, B=4 | ROUGE-1: 31.65 -> 0.0 | Improved |
| use SHE-LoRA instead of Baseline full HE | federated LoRA · OpenLLaMA-3B encryption time | mean time reduction: 99.87% | Improved |
| use SHE-LoRA instead of Baseline full HE | federated LoRA · OpenLLaMA-3B communication | communication overhead: 99.71% | Improved |
| encrypt the most important parameters first (Max strategy) instead of Min or Random (ablation) | SHE selection · mutual information with full parameters | mutual information | Improved |
| lower encryption budget below 0.3 per mille (ablation) | SHE-LoRA · DAGER | attack success | Hurt |
| increase batch size from 4 to 16 | Flex-LoRA · DAGER on SST2 | ROUGE-1: 95.18 -> 10.27 | Improved |
Closest papers by meaning
- Encryption-Friendly LLM Architecture (ICLR 2025 · similarity 0.89)
- Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning (NeurIPS 2026 · similarity 0.87)
- Differentially Private Federated Low Rank Adaptation Beyond Fixed-Matrix (NeurIPS 2025 · similarity 0.87)
- EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption (ICML 2025 · similarity 0.86)
- Improving LoRA in Privacy-preserving Federated Learning (ICLR 2024 · similarity 0.86)
- DictPFL: Efficient and Private Federated Learning on Encrypted Gradients (NeurIPS 2025 · similarity 0.85)
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.
