Research map / Architectures and efficiency / RNNs and normalization
Improving Neural Program Synthesis with Inferred Execution Traces
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 |
|---|---|---|---|
| insert inferred execution trace stage (I/O->TRACE->CODE) before program synthesis | Karel encoder-decoder synthesizer · Karel Top-1 Generalization | generalization accuracy: 73.5% -> 81.3% | Helped |
| insert inferred execution trace stage | Karel synthesizer · Karel Top-50 Guided Search | accuracy: 84.6% -> 88.8% | Helped |
| add trace stage, programs of length 30+ | Karel synthesizer · Karel programs length 30+ | accuracy: 48.6% -> 61.0% | Helped |
| add trace stage, short programs length 0-15 | Karel synthesizer · Karel programs length 0-15 | accuracy: 99.5% -> 99.5% | No clear effect |
| add interpreter grid state input to I/O->TRACE decoder (ablation) | I/O->TRACE model · Karel trace correctness Top-1 | trace correctness: 92.5% -> 94.8% | Helped |
| add interpreter grid state input to I/O->TRACE decoder (ablation) | I/O->TRACE model · Karel trace exact match Top-1 | exact match: 58.7% -> 57.6% | No clear effect |
| add residual connections to conv encoder (PRESNET) (ablation) | I/O->TRACE LGRL · Karel trace correctness Top-1 | trace correctness: 94.8% -> 95.2% | No clear effect |
| train TRACE->CODE on inferred traces instead of gold traces (ablation) | TRACE->CODE, tested on inferred traces · Karel Guided Search | accuracy: 81.8% -> 88.8% | Helped |
Closest papers by meaning
- Execution-Guided Neural Program Synthesis (ICLR 2019 · similarity 0.90)
- Synthesize, Execute and Debug: Learning to Repair for Neural Program Synthesis (NeurIPS 2020 · similarity 0.89)
- Towards Synthesizing Complex Programs From Input-Output Examples (ICLR 2018 · similarity 0.88)
- PLANS: Neuro-Symbolic Program Learning from Videos (NeurIPS 2020 · similarity 0.87)
- Program Synthesis via Test-Time Transduction (NeurIPS 2025 · similarity 0.85)
- ExeDec: Execution Decomposition for Compositional Generalization in Neural Program Synthesis (ICLR 2024 · similarity 0.85)
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.