Research map / Large language models / Agents, code and math
Learning Compositional Rules via Neural Program Synthesis
Maxwell Nye, Armando Solar-Lezama, Josh Tenenbaum, Brenden M Lake
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 |
|---|---|---|---|
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · SCAN length split | accuracy: 0.0 -> 100 | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · SCAN simple split | accuracy: 13.3 (3.3) -> 100 | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · SCAN add jump split | accuracy: 3.5 (0.7) -> 100 | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · SCAN add around right split | accuracy: 0.0 -> 100 | Improved |
| Use program synthesis with explicit checking rather than direct neural input-output prediction | Synth · SCAN length split | accuracy: 0.04 (0.02) -> 100 | Improved |
| Use program synthesis with explicit checking rather than direct neural input-output prediction | Synth · Spanish number-word learning | accuracy: 64.4 (3.2) -> 80.0 (17.9) | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · Spanish number-word learning | accuracy: 0.0 -> 80.0 (17.9) | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · French number-word learning | accuracy: 0.0 -> 75.5 (2.4) | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · Greek number-word learning | accuracy: 70.0 (10.2) -> 94.5 (4.9) | Improved |
| Add guess-and-check search to neural grammar synthesis (ablation) | Synth · Vietnamese number-word learning | accuracy: 69.5 (2.3) -> 69.5 (2.3) | No clear effect |
| Use search to find a support-consistent grammar (ablation) | Synth · MiniSCAN human instruction-learning query set | accuracy: 37.5% -> 98.75% | Improved |
| Use neural attention across support examples in place of separate example encoders | Neural program synthesis · MiniSCAN | accuracy: 3%, 4%, 3%, 3.5% vs. synthesis results | Improved |
Closest papers by meaning
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.88)
- Generating Pragmatic Examples to Train Neural Program Synthesizers (ICLR 2024 · similarity 0.87)
- Mastering Symbolic Operations: Augmenting Language Models with Compiled Neural Networks (ICLR 2024 · similarity 0.87)
- CraftFactory: A Conditioned Control Policy Benchmark for Compositional Generalization (AAAI 2025 · similarity 0.87)
- Rule Extrapolation in Language Modeling: A Study of Compositional Generalization on OOD Prompts (NeurIPS 2024 · similarity 0.86)
- Are Transformers Able to Reason by Connecting Separated Knowledge in Training Data? (ICLR 2025 · 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.
