Research map / Large language models / Retrieval and knowledge
Automated Formalization via Conceptual Retrieval-Augmented LLMs
Wangyue Lu, Lun Du, Sirui Li, Ke Weng, Haozhe Sun, Hengyu Liu, Minghe Yu, Tiancheng Zhang, Ge Yu
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 MCE module (ablation) | CRAMF · MiniF2F | FAR@10: 58.2% -> 63.1% | Improved |
| add DHR module (ablation) | CRAMF · MiniF2F | FAR@10: 48.5% -> 63.1% | Improved |
| add reranking (ablation) | CRAMF · MiniF2F | FAR@10: 54.3% -> 63.1% | Improved |
| add DHR module (ablation) | CRAMF · ProofNet | FAR@10: 36.8% -> 55.1% | Improved |
| add DHR module (ablation) | CRAMF · AdvancedMath | FAR@10: 37.1% -> 51.4% | Improved |
| add structured problem rewriting (ablation) | CRAMF · CombMath | FAR@10: 54.3% -> 63.1% | Improved |
| add CRAMF retrieval | Deepseek-V3 · AdvancedMath | FAR@10: 19.8% -> 32.1% | Improved |
| add CRAMF retrieval | Deepseek-V3 · MiniF2F | compilation pass rate@10: 52.3% -> 69.3% | Improved |
| add CRAMF retrieval | Claude 4.1 · AdvancedMath | compilation pass rate@10: 98.8% -> 99.4% | Improved |
| add CRAMF retrieval | Kimina-7B · MiniF2F | FAR@10: 80.3% -> 84.8% | Improved |
| replace LeanSearch retrieval with CRAMF | retrieval · MiniF2F | HitRate@3: 36.8% -> 44.2% | Improved |
| replace BM25 with CRAMF | retrieval · MiniF2F | HitRate@3: 11.3% -> 44.2% | Improved |
Closest papers by meaning
- DRIFT: Decompose, Retrieve, Illustrate, then Formalize Theorems (ICLR 2026 · similarity 0.93)
- Semantic Search over 9 Million Mathematical Theorems (NeurIPS 2026 · similarity 0.88)
- Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency (NeurIPS 2024 · similarity 0.88)
- ATLAS: Autoformalizing Theorems through Lifting, Augmentation, and Synthesis of Data (NeurIPS 2025 · similarity 0.88)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.86)
- LeanAgent: Lifelong Learning for Formal Theorem Proving (ICLR 2025 · 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.
