Research map / Label-efficient and robust learning / Domain adaptation and generalization
*-CFQ: Analyzing the Scalability of Machine Learning on a Compositional Task
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 |
|---|---|---|---|
| Increase training size while holding model size and training steps fixed | Transformer · *-CFQ random splits | error rate | Mixed |
| Expand the rule scope of the target dataset (ablation) | Transformer · *-CFQ random splits | error rate: roughly 3 times | Hurt |
| Add non-leaf rules rather than leaf rules (ablation) | Transformer · *-CFQ random splits | error rate | Hurt |
| Increase training size under compound divergence | Transformer · u-CFQ compound-divergence splits | error rate: an order of magnitude higher | Helped |
| Add equally weighted training examples from a related domain when in-domain data is scarce | Transformer · b-CFQ test set, 10k in-domain examples | error rate | Helped |
| Add equally weighted training examples from a related domain when in-domain training size is larger | Transformer · b-CFQ test set, 100k in-domain examples | error rate | Mixed |
| Increase the related-domain distance from the target domain | Transformer · b-CFQ test set, related-domain training | error rate | Hurt |
| Over-weight target-domain examples so at least half of observations are in-domain | Transformer · b-CFQ test set, related-domain training | error rate | Helped |
| Train on ungrounded rather than grounded examples (ablation) | Transformer · o-CFQ test set | accuracy | Hurt |
| Use substantially more ungrounded training examples | Transformer · o-CFQ test set | accuracy | Helped |
Closest papers by meaning
- Task Generalization with Autoregressive Compositional Structure: Can Learning from $D$ Tasks Generalize to $D^T$ Tasks? (ICML 2025 · similarity 0.87)
- metabench - A Sparse Benchmark of Reasoning and Knowledge in Large Language Models (ICLR 2025 · similarity 0.86)
- In-Context Compositional Learning vis Sparse Coding Transformer (NeurIPS 2025 · similarity 0.86)
- Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks (ICML 2024 · similarity 0.86)
- Limits of Transformer Language Models on Learning to Compose Algorithms (NeurIPS 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.