Research map / Label-efficient and robust learning / Recognition and long-tail
Characterizing Pattern Matching and Its Limits on Compositional Task Structures
Hoyeon Chang, Jinho Park, Hanseul Cho, Sohee Yang, Miyoung Ko, Hyeonbin Hwang, Seungpil Won, Dohaeng Lee, Youbin Ahn, Minjoon Seo
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 chain-of-thought supervision of intermediate states | GPT-2 on 3-Hop · 3-Hop | power-law exponent of required data: 2.58 -> 1.76 | Improved |
| add chain-of-thought supervision on Non-tree task | GPT-2 on Non-tree · Non-tree (|X|=50) | ID test accuracy | Mixed |
| scale GPT-2 from 68M to 1.5B parameters | GPT-2 on 2-Hop, Parallel 2-Hop, 3-Hop · scaling exponent | power-law exponent: 68M to 1.5B | No clear effect |
| scale GPT-2 to 1.5B parameters on Non-tree | GPT-2 on Non-tree · Non-tree ID test | ID test accuracy | No clear effect |
| replace GPT-2 with Mamba | 2-Hop scaling and Non-tree · 2-Hop / Non-tree | power-law exponent / ID generalization | No clear effect |
| increase training set size to near-exhaustive for Non-tree | GPT-2 on Non-tree · Non-tree (|X|=50) | ID test accuracy: 0.96 | Mixed |
| add partial computation examples (f1=f2) to training (ablation) | GPT-2 on 2-Hop · 2-Hop (|X|=50, N=10k) | logit-lens MRR of intermediate state: MRR of 0 in layers 3 to 8 at position x2 | Improved |
| increase k-cutoff evidence strength of test instances | GPT-2 on 2-Hop · 2-Hop (|X|=50, N=10k) | ID test accuracy | Improved |
Closest papers by meaning
- Scaling can lead to compositional generalization (NeurIPS 2025 · similarity 0.89)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.88)
- Slower Generalization, Faster Memorization: A Sweet Spot in Algorithmic Learning (NeurIPS 2026 · similarity 0.87)
- Task Generalization with Autoregressive Compositional Structure: Can Learning from $D$ Tasks Generalize to $D^T$ Tasks? (ICML 2025 · similarity 0.87)
- CraftFactory: A Conditioned Control Policy Benchmark for Compositional Generalization (AAAI 2025 · similarity 0.87)
- Does Data Scaling Lead to Visual Compositional Generalization? (ICML 2025 · similarity 0.87)
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.
