Research map / Large language models / Agents, code and math
Euclean: Automated Geometry Problem Formalization with Unified Verification in Lean
Linbin Tang, Jingyan You, Zilin Kang, Hanzhang Liu, Sophia Zhang, Zenan Li, Chenrui Cao, Liangcheng Song, Jiaao Wu, Xian Zhang, Fan Yang
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 concept formalization mapping to Mathlib primitives (ablation) | Euclean pipeline · OMNI-Geometry (30 problems) | compiling candidates (of 960): 125 -> 254 | Improved |
| add iterative compiler-feedback repair (ablation) | Euclean pipeline · OMNI-Geometry (30 problems) | compiling candidates (of 960): 254 -> 465 | Improved |
| add constraint formalization mapping (ablation) | Euclean pipeline · OMNI-Geometry (30 problems) | compiling candidates (of 960): 465 -> 500 | Improved |
| add Prove-First configuration anchoring (ablation) | Euclean pipeline · OMNI-Geometry (30 problems) | compiling candidates (of 960): 500 -> 480 | Hurt |
| add Prove-First configuration anchoring (ablation) | Euclean pipeline · OMNI-Geometry (30 problems, human Pass@3 paired) | semantic faithfulness (wins vs losses): 7 wins, 3 losses (10 both correct, 10 both incorrect) | Improved |
| sample five candidates (TOP5) instead of first only (TOP1) | Euclean · OMNI-Geometry (180 problems, difficulty 4.0+) | human-judged accuracy (%): 48.89 -> 73.33 | Improved |
| increase problem difficulty to 8.0+ | Euclean · OMNI-Geometry (human eval) | TOP1 accuracy (%): 50.94 -> 20.00 | Hurt |
| fine-tune Goedel v2 with SFT on proofs of Euclean-formalized geometry | Goedel v2 8B · Numina-Geometry | proof Pass@1 (%): 13.6 -> 15.1 | Improved |
| use DPO on pass/fail pairs instead of SFT (ablation) | Goedel v2 8B · Numina-Geometry | proof Pass@1 (%): 15.1 (SFT) vs 15.0 (DPO) | No clear effect |
| fine-tune DeepSeek-Prover-V2-7B with SFT on Euclean data | DeepSeek-Prover-V2-7B · Numina-Geometry | proof Pass@1: 13.4% -> 15.0% | Improved |
Closest papers by meaning
- Autoformalizing Euclidean Geometry (ICML 2024 · similarity 0.89)
- Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning (CVPR 2026 · similarity 0.88)
- Geoint-R1: Formalizing Multimodal Geometric Reasoning with Dynamic Auxiliary Constructions (CVPR 2026 · similarity 0.87)
- APOLLO: Automated LLM and Lean Collaboration for Advanced Formal Reasoning (NeurIPS 2025 · similarity 0.86)
- Numina-Lean-Agent: An Open and General Agentic Reasoning System for Formal Mathematics (ICML 2026 · similarity 0.86)
- Lean Workbook: A large-scale Lean problem set formalized from natural language math problems (NeurIPS 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.
