Research map / Reinforcement learning / Visual and representation RL
Learning to Guide a Saturation-Based Theorem Prover
Ibrahim Abdelaziz, Maxwell Crouse, Bassem Makni, Vernon Austel, Cristina Cornelio, Shajith Ikbal, Pavan Kapanipathi, Ndivhuwo Makondo, Kavitha Srinivas, Michael Witbrock, Achille Fokoue
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 StagedGCN vectorizer instead of vanilla GCN (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 998 -> 1,213 | Improved |
| use StagedGCN vectorizer instead of SAGE (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,150 -> 1,213 | Improved |
| use StagedGCN vectorizer instead of pattern-based (ablation) | TRAIL · M2k (cumulative) | problems solved: 1,765 -> 1,808 | Improved |
| use StagedGCN vectorizer instead of GCN in cross-dataset transfer (ablation) | TRAIL · M2k -> MPTP transfer | problems solved: 732 -> 1,048 | Improved |
| add layer normalization (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 799 -> 1,213 | Improved |
| add layer normalization (ablation) | TRAIL · M2k (cumulative) | problems solved: 1,462 -> 1,808 | Improved |
| add skip connections (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,129 -> 1,213 | Improved |
| use root node readout instead of averaging (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,212 -> 1,213 | No clear effect |
| embedding size 64 vs 512 (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,199 -> 1,213 | No clear effect |
| use time-based reward instead of step-based (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,192 -> 1,213 | Improved |
| add reward normalization (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,197 -> 1,213 | Improved |
| add reward bounding (ablation) | TRAIL · MPTP2078 (cumulative) | problems solved: 1,155 -> 1,213 | Improved |
Closest papers by meaning
- Reviving DSP for Advanced Theorem Proving in the Era of Reasoning Models (NeurIPS 2025 · similarity 0.89)
- ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis (ICML 2025 · similarity 0.89)
- LEGO-Prover: Neural Theorem Proving with Growing Libraries (ICLR 2024 · similarity 0.88)
- Learning Better Representations From Less Data For Propositional Satisfiability (NeurIPS 2024 · similarity 0.88)
- DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search (ICLR 2025 · similarity 0.87)
- miniCTX: Neural Theorem Proving with (Long-)Contexts (ICLR 2025 · similarity 0.87)
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.
