Research map / Large language models / Agents, code and math
Hypothesis-Driven Reasoning for Large Language Models
Aakash Kumar Agarwal, Moyuru Yamada
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 oracle hypotheses (semantic memory) instead of raw episodes only | GPT-4o · Experience-Transfer Lv.2 (no CoT) | accuracy (%): 35.3 -> 92.0 | Improved |
| add oracle semantic memory to raw episodes | Gemma3-12B · Experience-Transfer Lv.1 (no CoT) | accuracy (%): 18.0 -> 56.7 | Improved |
| add oracle semantic memory alongside episodic memory | GPT-4o · Experience-Transfer Lv.3 (CoT) | accuracy (%): 55.3 -> 66.0 (but semantic alone 68.7) | Mixed |
| replace per-episode Reflexion with HDR | HDR · Experience-Transfer Lv.2 (Gemma3-12B) | accuracy (%): 26.0 -> 80.0 | Improved |
| replace Reflexion (individual) with HDR | HDR · Experience-Transfer Lv.2 (GPT-4o) | accuracy (%): 59.3 -> 84.7 | Improved |
| HDR vs oracle hypotheses | HDR · Experience-Transfer Lv.2 (GPT-4o) | accuracy (%): 94.7 -> 84.7 | Hurt |
| add factor extraction to direct hypothesis (ablation) | HDR · Experience-Transfer Lv.2 | hypothesis F1: 0.0 -> 1.0 | Improved |
| add iterative refinement to simple two-stage (ablation) | HDR · Experience-Transfer Lv.3 | hypothesis F1: 0.36 -> 1.0 | Improved |
| use Gemini-2.5 Flash instead of GPT-4o for consolidation (ablation) | HDR · Experience-Transfer avg | accuracy (%): 72.2 -> 76.2 | Improved |
| use Gemma3-12B for consolidation (ablation) | HDR · Experience-Transfer Lv.1 | accuracy (%): 84.0 -> 90.7 (Lv.1), 84.7 -> 43.3 (Lv.2) | Mixed |
Closest papers by meaning
- Meaningful Learning: Enhancing Abstract Reasoning in Large Language Models via Generic Fact Guidance (NeurIPS 2024 · similarity 0.90)
- Training Language Models to Reason Efficiently (NeurIPS 2025 · similarity 0.90)
- Efficient Reasoning for Large Reasoning Language Models via Certainty-Guided Reflection Suppression (AAAI 2026 · similarity 0.89)
- ORACLE: Optimizing Reasoning Abilities of Large Language Models via Constraint-Led Synthetic Data Elicitation (AAAI 2026 · similarity 0.89)
- A State-Transition Framework for Efficient LLM Reasoning (ICLR 2026 · similarity 0.89)
- Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes (ICLR 2026 · similarity 0.89)
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.
