Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Compositional generalization through meta sequence-to-sequence learning
Brenden M Lake
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Train across episodes of changing sequence-to-sequence mappings and use a support memory | Meta seq2seq · SCAN add jump | test accuracy: 0.03% -> 99.95% | Improved |
| Train across episodes of changing sequence-to-sequence mappings and use a support memory | Meta seq2seq · SCAN around right | test accuracy: 0.0% -> 99.96% | Improved |
| Train across episodes of changing sequence-to-sequence mappings and use a support memory | Meta seq2seq · SCAN length | test accuracy: 7.71% -> 16.64% | Improved |
| Add support loss during meta-training (ablation) | Meta seq2seq · SCAN add jump | test accuracy: 5.43% -> 99.95% | Improved |
| Add support loss during meta-training (ablation) | Meta seq2seq · SCAN add jump, augmentation meta-training | test accuracy: 99.48% -> 98.71% | No clear effect |
| Add decoder attention over query context (ablation) | Meta seq2seq · SCAN add jump | test accuracy: 10.32% -> 99.95% | Improved |
| Add decoder attention over query context (ablation) | Meta seq2seq · SCAN add jump, augmentation meta-training | test accuracy: 9.29% -> 98.71% | Improved |
| Augment training with additional primitives | Standard seq2seq · SCAN add jump | test accuracy: 0.03% -> 12.26% | Improved |
| Use meta-training rather than standard training | Meta seq2seq · SCAN add jump, augmentation meta-training | test accuracy: 12.26% -> 98.71% | Improved |
| Use meta-training rather than standard training | Meta seq2seq · Mutual exclusivity | accuracy: 100% | Improved |
| Use meta-training rather than standard training | Meta seq2seq · SCAN around right | test accuracy: 0.0% -> 99.96% | Improved |
Closest papers by meaning
- Consistency of Compositional Generalization Across Multiple Levels (AAAI 2025 · similarity 0.87)
- Scaling can lead to compositional generalization (NeurIPS 2025 · similarity 0.85)
- Separating the 'what' and 'how' of compositional computation to enable reuse and continual learning (NeurIPS 2025 · similarity 0.85)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.84)
- Swing-by Dynamics in Concept Learning and Compositional Generalization (ICLR 2025 · similarity 0.84)
- Neural-Symbolic Recursive Machine for Systematic Generalization (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.
