Research map / Graph learning / Graph algorithms and clustering
Premise Selection for Theorem Proving by Deep Graph Embedding
Mingzhe Wang, Yihe Tang, Jian Wang, Jia Deng
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 |
|---|---|---|---|
| Use order-preserving graph updates instead of FormulaNet-basic updates | FormulaNet · HolStep test, conditional premise selection | accuracy: 89.1 -> 90.3 | Improved |
| Use graph-based formula embeddings rather than the strongest sequence baseline | HolStep · HolStep test, conditional premise selection | accuracy: 83 -> 90.3 | Improved |
| Use graph structure rather than an unstructured token bag (ablation) | FormulaNet-basic · HolStep validation, conditional premise selection | accuracy: 81.5 -> 89.9 | Improved |
| Use a parse tree instead of the variable-linked graph, with variables renamed (ablation) | FormulaNet-basic · Original Validation | accuracy: 89.9 -> 84.7 | Hurt |
| Use a parse tree instead of the variable-linked graph, with variables renamed (ablation) | FormulaNet-basic · Renamed Validation | accuracy: 89.9 -> 84.7 | Hurt |
| Retain original variable names instead of making the graph invariant to renaming (ablation) | FormulaNet-basic · Renamed Validation | accuracy: 89.9 -> 83.5 | Hurt |
| Retain original variable names instead of making the tree invariant to renaming (ablation) | FormulaNet-basic · Renamed Validation | accuracy: 89.9 -> 82.3 | Hurt |
| Increase FormulaNet-basic update depth from 0 to 3 steps (ablation) | FormulaNet-basic · HolStep validation, conditional premise selection | accuracy: 81.5 -> 89.9 | Improved |
| Increase FormulaNet update depth from 0 to 3 steps (ablation) | FormulaNet · HolStep validation, conditional premise selection | accuracy: 81.5 -> 91.1 | Improved |
| Increase FormulaNet update depth from 3 to 4 steps (ablation) | FormulaNet · HolStep validation, conditional premise selection | accuracy: 91.1 -> 90.8 | Hurt |
Closest papers by meaning
- Graph Representations for Higher-Order Logic and Theorem Proving (AAAI 2020 · similarity 0.90)
- Magnushammer: A Transformer-Based Approach to Premise Selection (ICLR 2024 · similarity 0.88)
- Improving Soft Unification with Knowledge Graph Embedding Methods (ICML 2025 · similarity 0.87)
- BAIT: Benchmarking (Embedding) Architectures for Interactive Theorem-Proving (AAAI 2024 · similarity 0.87)
- ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis (ICML 2025 · similarity 0.86)
- Reviving DSP for Advanced Theorem Proving in the Era of Reasoning Models (NeurIPS 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.
