Research map / Large language models / Agents, code and math
Synthesis of Programs from Multimodal Datasets
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 |
|---|---|---|---|
| use MultiSynth instead of baseline | baseline reordering · English-Farsi test | reordering accuracy: 85.17 -> 89.55 | Helped |
| use MultiSynth instead of baseline | baseline reordering · English-Italian test | reordering accuracy: 86.70 -> 90.35 | Helped |
| use MultiSynth instead of baseline | baseline reordering · English-Urdu test | reordering accuracy: 87.33 -> 90.65 | Helped |
| use MultiSynth instead of baseline on tree-transformable sentences | baseline · En-Fa | BLEU: 14.12 -> 77.51 | Helped |
| use MultiSynth instead of baseline on tree-transformable sentences | baseline · En-It | BLEU: 20.07 -> 79.84 | Helped |
| use MultiSynth instead of baseline on tree-transformable sentences | baseline · En-Ur | BLEU: 15.85 -> 65.35 | Helped |
| use MultiSynth instead of baseline on full shared-task data | baseline · En-Fa | BLEU: 50.0 -> 62.5 | Helped |
| use MultiSynth vs strongest prior system Vishwesariah et al. | Vishwesariah et al., 2011 · En-It | BLEU: 83.0 -> 73.8 | Hurt |
| use MultiSynth vs Dlougach and Galinskaya | Dlougach and Galinskaya, 2013 · En-Ur | BLEU: 55.8 -> 56.8 | Helped |
| learned ranking function for confused points vs baseline in XML domain | baseline · XML transformation S2-S5 | accuracy: +30% average | Helped |
| learned ranking under uncertainty (S5) | baseline · XML transformation S5 | accuracy: 36% average improvement | Helped |
| increase confused data from 20% to 60% (S2 to S4) | MultiSynth · XML P3 test | accuracy (%): 80 -> 53 | Hurt |
Closest papers by meaning
- HYSYNTH: Context-Free LLM Approximation for Guiding Program Synthesis (NeurIPS 2024 · similarity 0.86)
- It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs (NeurIPS 2026 · similarity 0.85)
- Efficient Multimodal Dataset Distillation via Generative Models (NeurIPS 2025 · similarity 0.84)
- OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling (ICML 2025 · similarity 0.84)
- Scalable Multilingual Multimodal Machine Translation with Speech-Text Fusion (ICLR 2026 · similarity 0.84)
- On Path to Multimodal Generalist: General-Level and General-Bench (ICML 2025 · 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.