Research map / Large language models / Agents, code and math
Coupling Distributed and Symbolic Execution for Natural Language Queries
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 |
|---|---|---|---|
| Couple distributed supervision with symbolic execution and REINFORCE | Symbolic executor · Overall | denotation accuracy (%): 75.8 -> 99.8 | Helped |
| Couple distributed supervision with symbolic execution and REINFORCE | Symbolic executor · Overall | execution accuracy (%): 49.5 -> 97.6 | Helped |
| Couple distributed supervision with symbolic execution and REINFORCE | Table QA · WhereSuperlative | denotation accuracy (%): 80.4 -> 99.9 | Helped |
| Couple distributed supervision with symbolic execution and REINFORCE | Table QA · NestQuery | denotation accuracy (%): 60.5 -> 100.0 | Helped |
| Feed symbolic intermediate results back to the distributed executor (ablation) | Distributed executor · Overall | accuracy (%): 84.0 -> 96.5 | Helped |
| Use execution labels for step-by-step training instead of denotation labels (ablation) | Distributed executor · Overall | accuracy (%): 84.0 -> 96.4 | Helped |
| Pretrain the symbolic executor's column predictors from distributed field attention (ablation) | Symbolic executor · Table QA | learning efficiency | Helped |
| Train the symbolic executor with REINFORCE alone instead of distributed-result pretraining (ablation) | Symbolic executor · Table QA | execution accuracy | Hurt |
| Feed imperfect intermediate field-attention labels from the distributed executor to symbolic column predictors (ablation) | Symbolic executor · Field selection | field selection accuracy: 70\% | Helped |
| Use symbolic execution instead of fully distributed execution | Table QA inference · Test set (10k samples), CPU | running time (seconds): 13.86 -> 2.65 | Helped |
| Use symbolic execution instead of fully distributed execution | Table QA inference · Test set (10k samples), GPU | running time (seconds): 1.05 -> 0.44 | Helped |
| Replace distributed row-selection gates with distributed row representations (ablation) | Distributed executor · Table QA | performance | Helped |
Closest papers by meaning
- SQL-R1: Training Natural Language to SQL Reasoning Model By Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Enhancing SQL Query Generation with Neurosymbolic Reasoning (AAAI 2025 · similarity 0.84)
- PARALLELPROMPT: Extracting Parallelism from Large Language Model Queries (NeurIPS 2025 · similarity 0.83)
- R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing (NeurIPS 2025 · similarity 0.83)
- NGDB-Zoo: Towards Efficient and Scalable Neural Graph Databases Training (NeurIPS 2026 · similarity 0.83)
- Peripheral Memory for LLMs: Integration of Sequential Memory Banks with Adaptive Querying (ICML 2025 · similarity 0.83)
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.