Research map / Label-efficient and robust learning / Recognition and long-tail
Learning to Recombine and Resample Data For Compositional Generalization
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 two-prototype recombination for augmentation rather than one-prototype recombination (ablation) | SCAN · jump and around right | accuracy: 0.17 \stderr{0.07} -> 0.87 \stderr{0.08} (around right); - -> 0.87 \stderr{0.08} (jump) | Helped |
| Add the copy mechanism to the recombination model (ablation) | Recomb-2 on SCAN · jump | accuracy: 0.00 \stderr{0.00} -> 0.88 \stderr{0.07} | Helped |
| Add a continuous latent variable (ablation) | Recomb-2 on SCAN · jump | accuracy: 0.88 \stderr{0.07} -> 0.88 \stderr{0.07} | No clear effect |
| Use recombination for data augmentation rather than direct inference (ablation) | Recomb-2 on SCAN · jump | accuracy: 0.57 \stderr{0.05} -> 0.88 \stderr{0.07} | Helped |
| Add resampling to two-prototype recombination (ablation) | Recomb-2 on SCAN · jump and around right | accuracy: 0.88 \stderr{0.07} -> 0.87 \stderr{0.08}; 0.82 \stderr{0.08} -> 0.75 \stderr{0.14} | Mixed |
| Add recombination-based synthetic examples rather than only resampling existing examples (ablation) | Morphological analysis · Swahili fut+pst | F1: 0.77 \stderr{0.01} -> 0.85 \stderr{0.01} | Helped |
| Add one-prototype recombination with resampling | Morphological analysis · fut+pst across Spanish, Swahili, and Turkish | F1: 0.66 -> 0.72 (Spanish); 0.75 -> 0.85 (Swahili); 0.69 -> 0.77 (Turkish) | Helped |
| Add one-prototype recombination with resampling | Morphological analysis · other word forms across Spanish, Swahili, and Turkish | F1: 0.88 -> 0.87 (Spanish); 0.90 -> 0.90 (Swahili); 0.85 -> 0.87 (Turkish) | Mixed |
| Combine GECA augmentation with recombination and resampling | Morphological analysis · Spanish and Turkish | F1: 0.72 -> 0.74 (Spanish fut+pst); 0.77 -> 0.79 (Turkish fut+pst) | Mixed |
Closest papers by meaning
- Compositional Generalization via Neural-Symbolic Stack Machines (NeurIPS 2020 · similarity 0.90)
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.89)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.87)
- Can Models Learn Skill Composition from Examples? (NeurIPS 2024 · similarity 0.86)
- Consistency of Compositional Generalization Across Multiple Levels (AAAI 2025 · similarity 0.86)
- Layer-Wise Representation Fusion for Compositional Generalization (AAAI 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.