Research map / Graph learning / GNN architectures
Graph Representations for Higher-Order Logic and Theorem Proving
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a 12-hop GNN with subexpression sharing instead of the WaveNet baseline | HOList proof search · HOList validation | proofs closed: 32.65% -> 49.95% | Helped |
| Use 12 message-passing hops rather than 0 hops (ablation) | Subexpression-sharing GNN · HOList validation | proofs closed: 40.86% -> 49.95% | Helped |
| Add subexpression sharing rather than use a plain AST (ablation) | 12-hop GNN · HOList validation | proofs closed: 45.67% -> 49.95% | Helped |
| Blind variable names (ablation) | Subexpression-sharing GNN · HOList validation | proofs closed: 49.95% -> 37.36% | Hurt |
| Add random edges (ablation) | Subexpression-sharing GNN · HOList validation | proofs closed: 49.95% -> 42.94% | Hurt |
| Restrict message passing to top-down edges instead of both directions (ablation) | Subexpression-sharing GNN · HOList validation | proofs closed: 49.95% -> 48.40% | Hurt |
| Restrict message passing to bottom-up edges instead of both directions (ablation) | Subexpression-sharing GNN · HOList validation | proofs closed: 49.95% -> 40.99% | Hurt |
| Share only identical leaves rather than full subexpressions (ablation) | GNN · HOList validation | proofs closed: 41.76% -> 29.24% (0 Hops to 4 Hops) | Hurt |
| Add variable blinding to leaf sharing (ablation) | GNN · HOList validation | proofs closed: 41.76% -> 31.78% (0 Hops); 29.24% -> 32.80% (4 Hops) | Mixed |
Closest papers by meaning
- ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis (ICML 2025 · similarity 0.88)
- Graph2Tac: Online Representation Learning of Formal Math Concepts (ICML 2024 · similarity 0.87)
- Improving Soft Unification with Knowledge Graph Embedding Methods (ICML 2025 · similarity 0.87)
- REFACTOR: Learning to Extract Theorems from Proofs (ICLR 2024 · similarity 0.86)
- Logical characterizations of recurrent graph neural networks with reals and floats (NeurIPS 2024 · similarity 0.86)
- LEGO-Prover: Neural Theorem Proving with Growing Libraries (ICLR 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.