Research map / Large language models / Agents, code and math
Abstracting Concept-Changing Rules for Solving Raven’s Progressive Matrix Problems
Fan Shi, Bin Li, Xiangyang Xue
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 |
|---|---|---|---|
| use CRAB concept-wise generation without auxiliary supervision instead of Transformer | CRAB · RAVEN average | selection accuracy: 59.8 -> 94.7 | Improved |
| use CRAB instead of Transformer on grid configuration | CRAB · RAVEN 3x3Grid | selection accuracy: 34.2 -> 90.2 | Improved |
| use CRAB instead of Transformer on inside-outside configuration | CRAB · RAVEN O-IC | selection accuracy: 14.5 -> 94.6 | Improved |
| use CRAB without auxiliary supervision instead of PrAE with distractors and rules | CRAB · RAVEN average | selection accuracy: 83.3 -> 94.7 | Improved |
| remove rule annotations from PrAE training (ablation) | PrAE · RAVEN average | selection accuracy: 83.3 -> 13.6 | Hurt |
| test on I-RAVEN instead of RAVEN, selective solver SCL | SCL · I-RAVEN vs RAVEN average | selection accuracy: 95.1 -> 85.7 | Hurt |
| test on I-RAVEN instead of RAVEN, CoPINet | CoPINet · I-RAVEN vs RAVEN average | selection accuracy: 95.6 -> 15.2 | Hurt |
| test on I-RAVEN instead of RAVEN, CRAB | CRAB · I-RAVEN vs RAVEN average | selection accuracy: 94.7 -> 95.9 | No clear effect |
| use CRAB instead of selective SCL on I-RAVEN grid | CRAB · I-RAVEN 3x3Grid | selection accuracy: 42.5 -> 91.1 | Improved |
| increase target images and candidate set size in arbitrary-position generation | CRAB · RAVEN GSA | global selection accuracy | No clear effect |
| increase target images from 1 to 2 or more | Transformer · RAVEN GSA | global selection accuracy | Hurt |
Closest papers by meaning
- Towards Generative Abstract Reasoning: Completing Raven’s Progressive Matrix via Rule Abstraction and Selection (ICLR 2024 · similarity 0.93)
- Learning to Disentangle Latent Reasoning Rules with Language VAEs: A Systematic Study (AAAI 2026 · similarity 0.86)
- GenVP: Generating Visual Puzzles with Contrastive Hierarchical VAEs (ICLR 2025 · similarity 0.85)
- Selection, Reflection and Self-Refinement: Revisit Reasoning Tasks via a Causal Lens (ICLR 2026 · similarity 0.85)
- Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning (NeurIPS 2026 · similarity 0.85)
- ActivationReasoning: Logical Reasoning in Latent Activation Spaces (ICLR 2026 · 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.
