Research map / Label-efficient and robust learning / Recognition and long-tail
Neuro-Symbolic Program Synthesis
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 |
|---|---|---|---|
| Replace linearized-tree io2seq generation with tree-structured R3NN generation | Program synthesis conditioned on I/O examples · Unseen program test set | accuracy: 42% -> 91% | Helped |
| Increase sampling from 1-best to 100 programs (ablation) | R3NN · Unseen program test set | accuracy: 63% -> 94% | Helped |
| Increase sampling from 100 to 300 programs (ablation) | R3NN · Unseen program test set | accuracy: 94% -> 97% | Helped |
| Use summed Cross Correlation instead of the baseline LSTM I/O encoder (ablation) | R3NN · Unseen program test set | accuracy: 88% -> 65% | Hurt |
| Use Diffused Cross Correlation instead of the baseline LSTM I/O encoder (ablation) | R3NN · Unseen program test set | accuracy: 88% -> 88% | No clear effect |
| Use LSTM-sum Cross Correlation instead of the baseline LSTM I/O encoder (ablation) | R3NN · Unseen program test set | accuracy: 88% -> 91% | Helped |
| Use Augmented Diffused Cross Correlation instead of the baseline LSTM I/O encoder (ablation) | R3NN · Unseen program test set | accuracy: 88% -> 91% | Helped |
| Train with 10 rather than 5 I/O examples per benchmark (ablation) | NSPS · 238 FlashFill benchmarks | tasks solved: 38.2% -> 29% | Hurt |
| Use pre-conditioning rather than root- or post-conditioning (ablation) | R3NN example encoding · Program synthesis | accuracy | Helped |
| Condition at all three tree locations instead of pre-conditioning alone (ablation) | R3NN example encoding · Program synthesis | accuracy | No clear effect |
Closest papers by meaning
- Gradient-Based Program Synthesis with Neurally Interpreted Languages (ICLR 2026 · similarity 0.90)
- RobustFill: Neural Program Learning under Noisy I/O (ICML 2017 · similarity 0.89)
- PLANS: Neuro-Symbolic Program Learning from Videos (NeurIPS 2020 · similarity 0.89)
- Towards Synthesizing Complex Programs From Input-Output Examples (ICLR 2018 · similarity 0.88)
- Learning to Compile Programs to Neural Networks (ICML 2024 · similarity 0.87)
- Generating Pragmatic Examples to Train Neural Program Synthesizers (ICLR 2024 · 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.