Research map / Large language models / Agents, code and math
AutoGPS: Automated Geometry Problem Solving via Multimodal Formalization and Deductive Reasoning
Bowen Ping, Minnan Luo, Zhuohang Dang, Chenxi Wang, Chengyou Jia
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 pre-formalization (diagram and text parsers) before MLLM formalization (ablation) | AutoGPS MPF · Geometry3K Completion | accuracy: +50.9% to +62.4% | Improved |
| add multimodal alignment to pre-formalization (ablation) | AutoGPS MPF · Geometry3K | accuracy: +7.8% to +13.0% | Improved |
| add multimodal alignment (ablation) | InterGPS solver · Geometry3K | accuracy: +4.2% | Improved |
| add minimal solution generation (ablation) | DSR · Geometry3K | average reasoning steps: 237.15 -> 16.69 | Improved |
| add geometry validation (ablation) | DSR · Geometry3K (ground-truth formalization) | Completion accuracy: 55.4 -> 94.5 | Improved |
| add geometry validation (ablation) | AutoGPS (GPT-4o) · Geometry3K | Completion accuracy: 55.5 -> 75.4 | Improved |
| use both deductive and algebraic reasoning (ablation) | DSR · Geometry3K (ground-truth formalization) | Completion accuracy: 5.5 -> 94.5 | Improved |
| use feedback-based Refine@5 instead of Major@5 (ablation) | AutoGPS · Geometry3K, PGPS9K, GeoQA Completion | accuracy: 66.3% -> 71.2% (Geo3K) | Improved |
| use feedback-based Refine@5 instead of Pass@5 (ablation) | AutoGPS · Geometry3K Completion | accuracy: 68.2% -> 71.2% | Improved |
| AutoGPS framework vs best neural/symbolic SOTA on Choice | AutoGPS (GPT-4o) · Geometry3K Choice | accuracy: 82.3 -> 81.6 | No clear effect |
| AutoGPS framework vs best SOTA on Choice | AutoGPS (GPT-4o) · PGPS9K Choice | accuracy: 73.8 -> 81.5 | Improved |
| AutoGPS framework vs best SOTA on Completion | AutoGPS (GPT-4o) · Geometry3K and PGPS9K Completion | accuracy: 71.3 -> 75.4; 66.1 -> 75.3 | Improved |
Closest papers by meaning
- Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning (CVPR 2026 · similarity 0.90)
- GeoBench: Rethinking Multimodal Geometric Problem-Solving via Hierarchical Evaluation (ICLR 2026 · similarity 0.90)
- Geoint-R1: Formalizing Multimodal Geometric Reasoning with Dynamic Auxiliary Constructions (CVPR 2026 · similarity 0.90)
- GNS: Solving Plane Geometry Problems by Neural-Symbolic Reasoning with Multi-Modal LLMs (AAAI 2025 · similarity 0.89)
- GeoX: Geometric Problem Solving Through Unified Formalized Vision-Language Pre-training (ICLR 2025 · similarity 0.89)
- SolidGeo: Measuring Multimodal Spatial Math Reasoning in Solid Geometry (NeurIPS 2025 · similarity 0.88)
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.
