Research map / Large language models / Agents, code and math
Learning Robust Rule Representations for Abstract Reasoning via Internal Inferences
Wenbo Zhang, Likai Tang, Site Mo, Xianggen Liu, Sen Song
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 discretization of rule representations (ablation) | ARII · PGM H.O.A.P. | accuracy (%): 47.5 -> 50.0 | Improved |
| add internal inference objective (ablation) | ARII · PGM H.O.A.P. | accuracy (%): 32.0 -> 50.0 | Improved |
| add both discretization and internal inference (ablation) | ARII · PGM H.O.A.P. | accuracy (%): 36.6 -> 50.0 | Improved |
| add discretization of rule representations (ablation) | ARII · PGM Interpolation | accuracy (%): 69.5 -> 72.0 | Improved |
| add internal inference objective (ablation) | ARII · PGM Interpolation | accuracy (%): 54.2 -> 72.0 | Improved |
| add both components (ablation) | ARII · PGM Interpolation | accuracy (%): 51.1 -> 72.0 | Improved |
| ARII instead of SCL on I-RAVEN | ARII · I-RAVEN Average | accuracy (%): 85.4 -> 91.1 | Improved |
| ARII instead of SCL on I-RAVEN 3x3 Grid | ARII · I-RAVEN 3x3 Grid | accuracy (%): 64.2 -> 78.9 | Improved |
| ARII instead of SCL on I-RAVEN U-D | ARII · I-RAVEN U-D | accuracy (%): 99.4 -> 90.3 | Hurt |
| ARII instead of MRNet on PGM H.O.A.P. | ARII · PGM H.O.A.P. | accuracy (%): 38.4 -> 50.0 | Improved |
| ARII instead of MRNet on PGM H.O.T.P. | ARII · PGM H.O.T.P. | accuracy (%): 55.3 -> 64.1 | Improved |
| ARII instead of MLRN on PGM Neutral | ARII · PGM Neutral | accuracy (%): 98.0 -> 88.0 | Hurt |
Closest papers by meaning
- Towards Generative Abstract Reasoning: Completing Raven’s Progressive Matrix via Rule Abstraction and Selection (ICLR 2024 · similarity 0.86)
- Benchmarking Abstract and Reasoning Abilities Through A Theoretical Perspective (ICML 2025 · similarity 0.86)
- PotARCin: Multi-Dimensional Evaluation of Skill Acquisition in Abstract Reasoning Tasks (NeurIPS 2026 · similarity 0.85)
- When can transformers reason with abstract symbols? (ICLR 2024 · similarity 0.85)
- Distinguishing Performance From Competence in Evaluations of Humanlike Abstract Reasoning (NeurIPS 2026 · similarity 0.85)
- Boule or Baguette? A Study on Task Topology, Length Generalization, and the Benefit of Reasoning Trace (NeurIPS 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.
