Research map / Large language models / Retrieval and knowledge
Pisces: Cryptography-based Private Retrieval-Augmented Generation with Dual-Path Retrieval
Xiaojian Liang, Lushan Song, Shishuai Du, Weicheng Zhu, Tan Li Hui Faith, Jun Jie Sim, Haibing Jin, Zhenghao Wu, Yingting Liu, Xin Zhang, Jiang-Ming Yang, Pu Duan
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 coarse-to-fine strategy (SimHash coarse matching) to encrypted semantic retrieval (ablation) | Pisces semantic path · large datasets (HotpotQA) | retrieval time: saves 38.78% ~ 41.21% time | Improved |
| add coarse-to-fine strategy to encrypted semantic retrieval (ablation) | Pisces semantic path · large datasets | upload overhead: reduces by 67.53% ~ 67.78% | Improved |
| add coarse-to-fine strategy to encrypted semantic retrieval (ablation) | Pisces semantic path · large datasets | download overhead: reduces by 68.49% ~ 68.77% | Improved |
| add coarse-to-fine strategy to encrypted semantic retrieval (ablation) | Pisces semantic path · ClapNQ Dev answerable (small) | retrieval time: 1.855 -> 3.56 | Hurt |
| add coarse-to-fine strategy to encrypted semantic retrieval (ablation) | Pisces semantic path · ClapNQ Dev answerable | retrieval accuracy: 100.00% -> 86.83% | Hurt |
| use multi-instance labeled PSI instead of labeled PSI (LSE) (ablation) | Pisces lexical path · HotpotQA Dev distractor | running time (s): 1051.98 -> 2.35 | Improved |
| use multi-instance labeled PSI instead of labeled PSI (LSE) (ablation) | Pisces lexical path · HotpotQA Dev distractor | upload (MB): 161.61 -> 0.0006 | Improved |
| use multi-instance labeled PSI instead of labeled PSI (LSE) (ablation) | Pisces lexical path · HotpotQA Dev distractor | download (MB): 382.26 -> 79.82 | Improved |
| combine lexical with semantic retrieval (dual path) instead of semantic only (ablation) | Pisces · ClapNQ Dev answerable | Top-5 accuracy vs ground truth: 30.14% -> 58.05% | Improved |
| use Pisces instead of DP-RAG (eps=1) | RAG retrieval · ClapNQ Dev answerable | Top-5 accuracy vs ground truth: 26.37% -> 58.05% | Improved |
| use Pisces instead of RemoteRAG (eps=1280) | RAG retrieval · ClapNQ Dev answerable | Top-5 accuracy vs ground truth: 35.54% -> 58.05% | Improved |
| use Pisces instead of plaintext retrieval | RAG retrieval · ClapNQ Dev answerable | Top-5 accuracy vs ground truth: 59.66% -> 58.05% | Mixed |
Closest papers by meaning
- InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy (NeurIPS 2025 · similarity 0.88)
- RAPID: Retrieval Augmented Training of Differentially Private Diffusion Models (ICLR 2025 · similarity 0.85)
- SeCon-RAG: A Two-Stage Semantic Filtering and Conflict-Free Framework for Trustworthy RAG (NeurIPS 2025 · similarity 0.85)
- Beyond Raw Context Transfer: Representation-based Federated Retrieval-Augmented Generation (NeurIPS 2026 · similarity 0.85)
- Encryption-Friendly LLM Architecture (ICLR 2025 · similarity 0.84)
- SynBench: A Benchmark for Differentially Private Text Generation (NeurIPS 2026 · 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.
