Research map / Label-efficient and robust learning / Recognition and long-tail
Compositional Generalization through Gradient Search in Nonparametric Latent Space
Haruki Shirakami, James Henderson
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 |
|---|---|---|---|
| add KL-regularization (ablation) | Abduction Transformer · 1-D ARC OOD composition | solve rate (%): 16.7 -> 25.1 | Improved |
| add KL-regularization (ablation) | Abduction Transformer · SRAVEN OOD composition (train on 1%) | solve rate (%): 16.8 -> 46.1 | Improved |
| add KL-regularization (ablation) | Abduction Transformer · SRAVEN OOD composition (train on 90%) | solve rate (%): 33.7 -> 96.4 | Improved |
| add test-time gradient search (ablation) | Abduction Transformer · 1-D ARC OOD composition | solve rate (%): 0.1 -> 25.1 | Improved |
| add test-time gradient search (ablation) | Abduction Transformer · SRAVEN OOD composition (train on 1%) | solve rate (%): 20.9 -> 46.1 | Improved |
| add test-time gradient search (ablation) | Abduction Transformer · SRAVEN OOD composition (train on 90%) | solve rate (%): 96.7 -> 96.4 | No clear effect |
| use nonparametric mixture latent instead of single-vector latent (LPN) | Abduction Transformer · 1-D ARC OOD composition | solve rate (%): 1.9 -> 25.1 | Improved |
| use nonparametric mixture latent instead of single-vector latent (LPN) | Abduction Transformer · SRAVEN OOD composition (train on 1%) | solve rate (%): 37.1 -> 46.1 | Improved |
| use nonparametric mixture latent instead of single-vector latent (LPN) | Abduction Transformer · SRAVEN OOD composition (train on 90%) | solve rate (%): 93.5 -> 96.4 | Improved |
| stochastic Dirichlet-process encoder instead of deterministic encoder (ablation) | Abduction Transformer · 1-D ARC OOD composition | solve rate (%): 0.1 -> 25.1 | Improved |
| stochastic Dirichlet-process encoder instead of deterministic encoder (ablation) | Abduction Transformer · SRAVEN OOD composition (train on 1%) | solve rate (%): 10.8 -> 46.1 | Improved |
| stochastic Dirichlet-process encoder instead of deterministic encoder (ablation) | Abduction Transformer · SRAVEN OOD composition (train on 90%) | solve rate (%): 27.2 -> 96.4 | Improved |
Closest papers by meaning
- In-Context Compositional Learning vis Sparse Coding Transformer (NeurIPS 2025 · similarity 0.87)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.87)
- Scaling can lead to compositional generalization (NeurIPS 2025 · similarity 0.87)
- Scalable Evaluation and Neural Models for Compositional Generalization (NeurIPS 2025 · similarity 0.87)
- Layer-Wise Representation Fusion for Compositional Generalization (AAAI 2024 · similarity 0.85)
- Consistency of Compositional Generalization Across Multiple Levels (AAAI 2025 · similarity 0.85)
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.
