Research map / Large language models / Language, safety and interpretability
Abstract Reasoning with Distracting Features
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the reinforcement-learning teacher's adaptive training trajectory | LEN · PGM neutral train/test split | accuracy (%): 68.1 -> 79.8 | Helped |
| add the reinforcement-learning teacher's adaptive training trajectory | LEN · RAVEN | accuracy (%): 72.9 -> 78.3 | Helped |
| add type loss (ablation) | LEN · PGM neutral train/test split | accuracy (%): 68.1 -> 82.3 | Helped |
| add type loss (ablation) | LEN · RAVEN | accuracy (%): 72.9 -> 59.4 | Hurt |
| add the two-stream architecture (ablation) | LEN · PGM neutral train/test split | accuracy (%): 68.1 -> 70.3 | Helped |
| add type loss to the teacher-guided T-LEN (ablation) | T-LEN + Teacher Model · PGM neutral train/test split | accuracy (%): 85.1 -> 88.9 | Helped |
| use the teacher model rather than random training order | LEN on datasets 3 and 4 with distracting features · Trajectory analysis | accuracy (%): 58.2 -> 67.6 | Helped |
| use the teacher model rather than random training order | LEN on datasets 1 and 2 without distracting features · Trajectory analysis | accuracy (%): 81.0 -> 82.1 | Helped |
| add self-paced learning (ablation) | LEN · PGM neutral train/test split | accuracy (%): 68.1 -> 57.2 | Hurt |
| add curriculum learning (ablation) | LEN · PGM neutral train/test split | accuracy (%): 68.1 -> 63.3 | Hurt |
| add focal loss (ablation) | LEN · RAVEN | accuracy (%): 72.9 -> 75.6 | Helped |
| add Mentornet-PD (ablation) | LEN · RAVEN | accuracy (%): 72.9 -> 74.4 | Helped |
Closest papers by meaning
- Slot Abstractors: Toward Scalable Abstract Visual Reasoning (ICML 2024 · similarity 0.82)
- Distinguishing Performance From Competence in Evaluations of Humanlike Abstract Reasoning (NeurIPS 2026 · similarity 0.82)
- Benchmarking Abstract and Reasoning Abilities Through A Theoretical Perspective (ICML 2025 · similarity 0.82)
- Exposing Weaknesses of Large Reasoning Models through Graph Algorithm Problems (ICLR 2026 · similarity 0.81)
- Answering the Unanswerable Is to Err Knowingly: Analyzing and Mitigating Abstention Failures in Large Reasoning Models (AAAI 2026 · similarity 0.80)
- PotARCin: Multi-Dimensional Evaluation of Skill Acquisition in Abstract Reasoning Tasks (NeurIPS 2026 · similarity 0.80)
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.