Research map / Large language models / Retrieval and knowledge
MedCoG: Maximizing LLM Inference Density in Medical Reasoning via Meta-Cognitive Regulation
Yu Zhao, Hao Guan, Yongcheng Jing, Ying Zhang, Dacheng Tao
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 meta-cognition regulation (on-demand strategy selection) instead of always using all three knowledge types (ablation) | MedCoG · Hard subsets average | accuracy: 33.2 -> 37.5 | Improved |
| add meta-cognition regulation instead of always using all knowledge types (ablation) | MedCoG · IIE per 1k samples | IIE: 0.181 -> 0.438 | Improved |
| add KG retrieval to structured CoT (ablation) | SCoT (GPT-4o) · MedQA-Hard | accuracy: 41.0 -> 37.0 | Hurt |
| add KG retrieval to structured CoT (ablation) | SCoT (GPT-4o) · MedQA-Full | accuracy: 89.55 -> 87.43 | Hurt |
| add episodic memory to structured CoT (ablation) | SCoT (GPT-4o) · MedQA-Hard | accuracy: 41.0 -> 42.0 | No clear effect |
| add both KG and memory to structured CoT (ablation) | SCoT (GPT-4o) · MedQA-Hard | accuracy: 41.0 -> 50.0 | Improved |
| add memory to KG-augmented SCoT (ablation) | SCoT+KG (GPT-4o) · MedQA-Hard | accuracy: 37.0 -> 50.0 | Improved |
| replace always-on strategy with oracle strategy selection | MedCoG · MedQA-Hard | accuracy: 50.0 -> 67.0 | Improved |
| add per-backbone threshold calibration (ablation) | MedCoG monitor GPT-4o-mini · MedQA-Hard | accuracy: 40 -> 50 | Improved |
| add per-backbone threshold calibration (ablation) | MedCoG monitor o3-mini · MedQA-Hard | accuracy: 44 -> 49 | Improved |
| add per-backbone threshold calibration (ablation) | MedCoG monitor Gemini-2.0-Flash · MedQA-Hard | accuracy: 48 -> 52 | Improved |
| add per-backbone threshold calibration (ablation) | MedCoG monitor GPT-4o · MedQA-Hard | accuracy: 52 -> 52 | No clear effect |
Closest papers by meaning
- Evaluating LLMs Across Multi-Cognitive Levels: From Medical Knowledge Mastery to Scenario-Based Problem Solving (ICML 2025 · similarity 0.90)
- MedS³: Towards Medical Slow Thinking with Self-Evolved Soft Dual-sided Process Supervision (AAAI 2026 · similarity 0.88)
- Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks (NeurIPS 2026 · similarity 0.87)
- Optimizing Inference-Time Compute for Medical Reasoning via Uncertainty Quantification (ICML 2026 · similarity 0.87)
- Simulating Viva Voce Examinations to Evaluate Clinical Reasoning in Large Language Models (NeurIPS 2025 · similarity 0.87)
- MR-Ben: A Meta-Reasoning Benchmark for Evaluating System-2 Thinking in LLMs (NeurIPS 2024 · similarity 0.87)
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.
