Research map / Reinforcement learning / Visual and representation RL
Learning to Solve SMT Formulas
Mislav Balunovic, Pavol Bielik, Martin Vechev
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 |
|---|---|---|---|
| synthesize a single branching strategy instead of using best strategy per formula (ablation) | fastSMT · SMT-LIB benchmarks | % formulas solved only by learned: 16.8% -> 8.9% | Hurt |
| use learned strategies instead of Z3 default strategy | Z3 · SMT-LIB benchmarks | formulas solved only by learned: 16.8% vs 0.3% | Improved |
| use synthesized combined strategy instead of Z3 default | Z3 · SMT-LIB benchmarks | formulas solved: 8.9% vs 3.6% solved only by one | Improved |
| use neural network policy instead of bilinear model, BFS, random or evolutionary search (ablation) | fastSMT policy · SMT-LIB test set | relative best strategy score | Improved |
| replace learned search with evolutionary algorithm (ablation) | fastSMT policy · SMT-LIB test set | relative best strategy score | No clear effect |
| represent formula with bag-of-words rather than skip-gram embeddings (ablation) | NN policy · SMT-LIB test set | relative best strategy score | Improved |
| represent formula with AST subtree features instead of bag-of-words (ablation) | NN policy · SMT-LIB test set | relative best strategy score | Hurt |
| add formula features (probes, BOW) to strategy-only input (ablation) | NN policy · SMT-LIB test set | relative best strategy score | Mixed |
| increase evaluation time limit from 10s to 600s | fastSMT · SMT-LIB benchmarks | % formulas solved: 85.1% -> 97.7% | Improved |
Closest papers by meaning
- Composing Biases by Using CP to Decompose Minimal Functional Dependencies for Acquiring Complex Formulae (AAAI 2024 · similarity 0.85)
- Improving Monte Carlo Tree Search for Symbolic Regression (NeurIPS 2025 · similarity 0.84)
- rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking (ICML 2025 · similarity 0.84)
- Teaching Transformers to Solve Combinatorial Problems through Efficient Trial & Error (NeurIPS 2025 · similarity 0.84)
- SolverLLM: Leveraging Test-Time Scaling for Optimization Problem via LLM-Guided Search (NeurIPS 2025 · similarity 0.84)
- JiuZhang3.0: Efficiently Improving Mathematical Reasoning by Training Small Data Synthesis Models (NeurIPS 2024 · similarity 0.84)
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.
