Research map / Large language models / Retrieval and knowledge
Mnemosyne: Accelerating Multi-Hop Question Answering via Cache Hit Order Fitting
Haizhou Du, Jiujiu Li, Dongyang Li, Luobin Huang, Lisheng Wang
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 cache-aware order fitting (graph reordering of queries) (ablation) | Mnemosyne · HotpotQA (sparse) | cache hit rate: 22.9 -> 35.3 | Improved |
| add cache-aware order fitting (ablation) | Mnemosyne · HotpotQA (sparse) | latency (s): 0.28 -> 0.18 | Improved |
| add L1 exact-match cache (ablation) | Mnemosyne · HotpotQA (sparse) | F1: 54.2 -> 54.6 | Improved |
| add L2 entity-document cache (ablation) | Mnemosyne · HotpotQA (sparse) | cache hit rate: 4.1 -> 35.3 | Improved |
| add idempotency rule (ablation) | Mnemosyne · HotpotQA (sparse) | F1: 54.0 -> 54.6 | Improved |
| raise cache tolerance to 0.4 (ablation) | Mnemosyne · HotpotQA (sparse) | F1: 54.6 -> 42.4 | Hurt |
| lower cache tolerance to 0.1 (ablation) | Mnemosyne · HotpotQA (sparse) | F1: 54.6 -> 48.9 | Hurt |
| increase cache capacity from 20 to 40 (ablation) | Mnemosyne · HotpotQA (sparse) | F1: 54.6 -> 55.7 | Improved |
| increase cache capacity from 40 to 80 (ablation) | Mnemosyne · HotpotQA (sparse) | F1: 55.7 -> 55.9 | No clear effect |
| use approximate semantic (Proximity) caching instead of entity-level caching | RAG caching baseline · HotpotQA (exact dense) | delta answer F1: -13.6 | Hurt |
| exact-match query caching (CAG-R) | naive RAG · four MHQA datasets | retrieval speedup: 0.87-0.98x | Hurt |
Closest papers by meaning
- Cooperative Retrieval-Augmented Generation for Question Answering: Mutual Information Exchange and Ranking by Contrasting Layers (NeurIPS 2025 · similarity 0.89)
- Autofocus Retrieval: An Effective Pipeline for Multi-Hop Question Answering With Semi-Structured Knowledge (NeurIPS 2026 · similarity 0.88)
- Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models (AAAI 2024 · similarity 0.88)
- Knowledge Editing with Dynamic Knowledge Graphs for Multi-Hop Question Answering (AAAI 2025 · similarity 0.87)
- MEQA: A Benchmark for Multi-hop Event-centric Question Answering with Explanations (NeurIPS 2024 · similarity 0.86)
- Question Calibration and Multi-Hop Modeling for Temporal Question Answering (AAAI 2024 · similarity 0.86)
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.
