Research map / Large language models / Language, safety and interpretability
Measuring Compositional Generalization: A Comprehensive Method on Realistic Data
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 |
|---|---|---|---|
| Evaluate LSTM+attention on an MCD split instead of a random split | LSTM+attention · CFQ | accuracy (%): 97.4 $\pm$0.3 -> 14.9 $\pm$1.1 | Hurt |
| Evaluate Transformer on an MCD split instead of a random split | Transformer · CFQ | accuracy (%): 98.5 $\pm$0.2 -> 17.9 $\pm$0.9 | Hurt |
| Evaluate Universal Transformer on an MCD split instead of a random split | Universal Transformer · CFQ | accuracy (%): 98.0 $\pm$0.3 -> 18.9 $\pm$1.4 | Hurt |
| Evaluate LSTM+attention on an MCD split instead of a random split | LSTM+attention · SCAN | accuracy (%): 99.9 $\pm$2.7 -> 6.1 $\pm$2.2 | Hurt |
| Evaluate Transformer on an MCD split instead of a random split | Transformer · SCAN | accuracy (%): 100.0 $\pm$0.0 -> 1.1 $\pm$0.5 | Hurt |
| Evaluate Universal Transformer on an MCD split instead of a random split | Universal Transformer · SCAN | accuracy (%): 99.9 $\pm$0.2 -> 1.2 $\pm$0.7 | Hurt |
| Use MCD splits rather than random, length-based, or pattern-based splits | DBCA split construction · CFQ | compound divergence: Random 0.000; Output Length 0.176; Input Length 0.062; Output Pattern 0.008; Input Pattern 0.005; MCD 0.694, 0.713, 0.70 | Helped |
| Use MCD splits rather than random, length-based, or pattern-based splits | DBCA split construction · SCAN | compound divergence: Random 0.047; Output Length 0.437; Input Length 0.380; Output Pattern 0.221; Input Pattern 0.240; MCD 0.736, 0.734, 0.73 | Helped |
Closest papers by meaning
- Consistency of Compositional Generalization Across Multiple Levels (AAAI 2025 · similarity 0.91)
- Hierarchical Poset Decoding for Compositional Generalization in Language (NeurIPS 2020 · similarity 0.91)
- *-CFQ: Analyzing the Scalability of Machine Learning on a Compositional Task (AAAI 2021 · similarity 0.90)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.88)
- Compositional Generalization for Multi-Label Text Classification: A Data-Augmentation Approach (AAAI 2024 · similarity 0.86)
- Scalable Evaluation and Neural Models for Compositional Generalization (NeurIPS 2025 · 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.