Research map / Large language models / Agents, code and math
Non-Parametric Structural Priors for Geometry Theorem Prediction
Junbo Zhao, Ting Zhang, Can Li, Wei He, Jingdong Wang, Hua Huang
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 RAG-retrieved solver-validated candidate theorems to Vanilla ICL (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 26.29 -> 72.64 | Improved |
| add TPG precedence prior on top of RAG (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 72.64 -> 84.42 | Improved |
| add TPG on hard problems (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) Hard | accuracy (%): 22.95 -> 40.98 | Improved |
| add global prior to Vanilla ICL (ablation) | Vanilla ICL · FormalGeo7K (1400 test) | accuracy (%): 26.29 -> 29.71 | Improved |
| use query-adaptive prior instead of global prior (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 29.71 -> 58.43 | Improved |
| use state-aware prior with symbolic validation instead of query-adaptive prior (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 58.43 -> 84.42 | Improved |
| add iterative symbolic feedback (vs single-pass) (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 34.3 -> 84.42 | Improved |
| increase retrieval pool Top-K from 15 to 200 (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 71.86 -> 84.42 | Improved |
| increase Top-K from 30 to 100 (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) | accuracy (%): 79.00 -> 80.29 | Improved |
| increase Top-K from 100 to 200 on Hard (ablation) | Pri-TPG (GPT-5 mini) · FormalGeo7K (1400 test) Hard | accuracy (%): 30.33 -> 40.98 | Improved |
| swap backbone from GPT-5 mini to GPT-5.2 in Pri-TPG | Pri-TPG · FormalGeo7K (1400 test) | accuracy (%): 84.42 -> 89.29 | Improved |
| wrap GPT-5.2 in Pri-TPG versus direct solving | GPT-5.2 · FormalGeo7K (1400 test) | accuracy (%): 73.14 -> 89.29 | Improved |
Closest papers by meaning
- miniCTX: Neural Theorem Proving with (Long-)Contexts (ICLR 2025 · similarity 0.86)
- Scaling-Aware Adapter for Structure-Grounded LLM Reasoning (ICML 2026 · similarity 0.86)
- Topology of Reasoning: Understanding Large Reasoning Models through Reasoning Graph Properties (NeurIPS 2025 · similarity 0.85)
- Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning (CVPR 2026 · similarity 0.85)
- DAG-Math: Graph-Guided Mathematical Reasoning in LLMs (ICLR 2026 · similarity 0.85)
- Geometry Awakening: Cross-Geometry Learning Exhibits Superiority over Individual Structures (NeurIPS 2024 · similarity 0.85)
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.
