Research map / Architectures and efficiency / Attention and transformers
Latent Attention For If-Then 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 |
|---|---|---|---|
| ensemble of 10 joint models by voting (ablation) | Our model · IFTTT joint all four attributes | top-1 accuracy: 43.3% -> 44.5% | Helped |
| ensemble of 10 joint models by voting (ablation) | Our model · IFTTT TC+TF | top-1 accuracy: 61.3% -> 62.7% | Helped |
| ensemble of 10 joint models by voting (ablation) | Our model · IFTTT all four attributes | top-5 accuracy: 73.3% -> 76.4% | Helped |
| joint instead of separate training (ablation) | Our model · IFTTT all four attributes | top-1 accuracy: 42.4% -> 43.3% | Helped |
| joint instead of separate training (ablation) | Our model · IFTTT AC+AF | top-1 accuracy: 62.0% -> 61.7% | Hurt |
| joint instead of separate training (ablation) | Our model · IFTTT TC+TF | top-1 accuracy: 61.2% -> 61.3% | No clear effect |
| latent attention model instead of Memn2n-2 (joint) | Our model · IFTTT all four attributes | top-1 accuracy: 41.9% -> 43.3% | Helped |
| our model instead of Seq2seq | Our model · IFTTT all four attributes | top-1 accuracy: 25.1% -> 43.3% | Helped |
| our model vote vs oracleturk | Our model · IFTTT TC+AC | top-1 accuracy: 56.0% -> 56.6% | Helped |
| our model instead of Memn2n-2 on rare categories (part b) | Our model · IFTTT 18k unbalanced trigger channel | top-1 accuracy: 33.1% -> 40.5% | Helped |
| our model instead of Memn2n-2 on all categories | Our model · IFTTT 18k unbalanced trigger channel | top-1 accuracy: 72.0% -> 73.2% | Helped |
| our model instead of Memn2n-2 on frequent categories (part a) | Our model · IFTTT 18k unbalanced trigger channel | top-1 accuracy: 79.4% -> 79.4% | No clear effect |
Closest papers by meaning
- Latent Programmer: Discrete Latent Codes for Program Synthesis (ICML 2021 · similarity 0.89)
- Leveraging Grammar and Reinforcement Learning for Neural Program Synthesis (ICLR 2018 · similarity 0.86)
- Searching Latent Program Spaces (NeurIPS 2025 · similarity 0.85)
- HYSYNTH: Context-Free LLM Approximation for Guiding Program Synthesis (NeurIPS 2024 · similarity 0.85)
- Generating Pragmatic Examples to Train Neural Program Synthesizers (ICLR 2024 · similarity 0.85)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 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.