Research map / Large language models / Agents, code and math
GamePad: A Learning Environment for 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 learned proof-state embeddings with a GRU or TreeLSTM instead of heuristic SVM features | Feit-Thompson position evaluation · Feit-Thompson | position evaluation accuracy (kernel-level): 57.37 -> 68.44 | Helped |
| Use learned proof-state embeddings with a GRU or TreeLSTM instead of heuristic SVM features | Feit-Thompson tactic prediction · Feit-Thompson | tactic prediction accuracy (kernel-level): 48.94 -> 60.63 | Helped |
| Remove implicit arguments by using mid-level proof states (ablation) | Feit-Thompson position evaluation · Feit-Thompson | position evaluation accuracy: 68.44 -> 66.30 | No clear effect |
| Remove implicit arguments by using mid-level proof states (ablation) | Feit-Thompson tactic prediction · Feit-Thompson | tactic prediction accuracy: 60.63 -> 60.55 | No clear effect |
| Add embedding sharing | GRU position evaluation training · Feit-Thompson | training time speedup: 1 -> 10× | Helped |
| Add dynamic batching | GRU position evaluation training · Feit-Thompson | training time speedup: 1 -> 10× | Helped |
| Add embedding sharing and dynamic batching | GRU position evaluation training · Feit-Thompson | training time speedup: 1 -> 130× | Helped |
| Use a greedy learned tactic predictor without backtracking | Algebraic rewrite proof synthesis · Algebraic rewrites, expression length 10 | complete proofs: 14 of 50 | Mixed |
| Use deterministic proof steps when the learned step fails | Algebraic rewrite proof synthesis · Algebraic rewrites, expression length 10 | complete proofs: 14 of 50 -> all 50 | Helped |
Closest papers by meaning
- BAIT: Benchmarking (Embedding) Architectures for Interactive Theorem-Proving (AAAI 2024 · similarity 0.88)
- Reviving DSP for Advanced Theorem Proving in the Era of Reasoning Models (NeurIPS 2025 · similarity 0.88)
- MPS-Prover: Advancing Stepwise Theorem Proving by Multi-Perspective Search and Data Curation (NeurIPS 2025 · similarity 0.88)
- Subgoal-based Demonstration Learning for Formal Theorem Proving (ICML 2024 · similarity 0.87)
- REFACTOR: Learning to Extract Theorems from Proofs (ICLR 2024 · similarity 0.87)
- Magnushammer: A Transformer-Based Approach to Premise Selection (ICLR 2024 · 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.