Research map / Vision-language and multimodal / Human and embodied modeling
Learning Algebraic Representation for Systematic Generalization in Abstract Reasoning
Chi Zhang, Sirui Xie, Baoxiong Jia, Ying Nian Wu, Song-Chun Zhu, Yixin Zhu
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 Peano-based algebraic matrix representation instead of independent encodings (ablation) | ALANS · RPM systematicity (Zhang data) | accuracy: 52.70% -> 78.45% | Improved |
| use algebraic representation instead of independent encodings (ablation) | ALANS · RPM productivity (Zhang data) | accuracy: 36.45% -> 79.95% | Improved |
| use algebraic representation instead of independent encodings (ablation) | ALANS · RPM localism (Zhang data) | accuracy: 59.80% -> 80.50% | Improved |
| use algebraic representation instead of independent encodings (ablation) | ALANS · RPM average (Hu data) | accuracy: 45.20% -> 65.42% | Improved |
| provide perfect visual annotations (perception) | ALANS · RPM average (Zhang data) | accuracy: 79.63% -> 93.12% | Improved |
| provide perfect visual annotations (perception) | ALANS · RPM average (Hu data) | accuracy: 65.42% -> 87.45% | Improved |
| use ALANS instead of CoPINet on systematic generalization | ALANS vs CoPINet · RPM average (Zhang data) | accuracy: 59.80% -> 79.63% | Improved |
| use ALANS instead of CoPINet on systematic generalization | ALANS vs CoPINet · RPM average (Hu data) | accuracy: 26.43% -> 65.42% | Improved |
| remove shortcut solutions (RAVEN to I-RAVEN) | CoPINet · RAVEN vs I-RAVEN IID | accuracy: 91.4% -> 46.1% | Hurt |
| remove shortcut solutions (RAVEN to I-RAVEN) | ALANS · RAVEN vs I-RAVEN IID | accuracy: 74.4% -> 78.5% | No clear effect |
| compare ALANS with CoPINet on RAVEN IID | ALANS · RAVEN IID | accuracy: 91.4% vs 74.4% | Hurt |
Closest papers by meaning
- Slot Abstractors: Toward Scalable Abstract Visual Reasoning (ICML 2024 · similarity 0.86)
- Think Visually, Reason Textually: Vision-Language Synergy in Abstract Reasoning (CVPR 2026 · similarity 0.86)
- Emergent Symbolic Mechanisms Support Abstract Reasoning in Large Language Models (ICML 2025 · similarity 0.86)
- CogFlow: Bridging Perception and Reasoning through Knowledge Internalization for Visual Mathematical Problem Solving (ICLR 2026 · similarity 0.85)
- How do Transformers Learn Implicit Reasoning? (NeurIPS 2025 · similarity 0.85)
- Composing Global Solutions to Reasoning Tasks via Algebraic Objects in Neural Nets (NeurIPS 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.
