Research map / Large language models / Reasoning and chain of thought
CARE: Causally-Aligned Reasoning Exploration for Medical Large Language Models
Yucheng Zhou, Peng Luo, Qianning Wang, Cheng-zhong Xu, Shen Jianbing
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 |
|---|---|---|---|
| use standard outcome-only GRPO on Hulu-Med-7B (ablation) | Hulu-Med-7B · avg of PMC-VQA, MedQA, MMMU-Med | average accuracy (%): 63.9 -> 64.3 | Improved |
| add causal sufficiency (agreement self-verification) (ablation) | CARE · avg of PMC-VQA, MedQA, MMMU-Med | average accuracy (%): 64.6 -> 65.4 | Improved |
| add proximal learnability window (ablation) | CARE · avg of PMC-VQA, MedQA, MMMU-Med | average accuracy (%): 64.9 -> 65.4 | Improved |
| add difficulty-weighted experience replay (ablation) | CARE · avg of PMC-VQA, MedQA, MMMU-Med | average accuracy (%): 65.1 -> 65.4 | Improved |
| full CARE vs standard GRPO | Hulu-Med-7B · avg of PMC-VQA, MedQA, MMMU-Med | average accuracy (%): 64.3 -> 65.4 | Improved |
| train only on proximal-window experiences (ablation) | CARE · accuracy gain by NLL region | accuracy gain (%): +5.2 | Improved |
| train only on trivial experiences (ablation) | CARE · accuracy gain by NLL region | accuracy gain (%): +0.4 | No clear effect |
| train only on extreme-hard experiences (ablation) | CARE · accuracy gain by NLL region | accuracy gain (%): -1.2 | Hurt |
| use CARE instead of standard GRPO for rationale validity | Hulu-Med-7B · MedQA (500 samples correct for both) | valid causal reasoning (%): Standard GRPO 45.0% spurious -> CARE 86.2% valid | Improved |
| use CARE for robustness to injected distractors | Hulu-Med-7B · MedQA with injected distractors | performance drop (%): 15.3 -> 3.8 | Improved |
| apply CARE to HuatuoGPT-V 7B | HuatuoGPT-V 7B · PMC-VQA | accuracy (%): 53.1 -> 54.6 | Improved |
Closest papers by meaning
- MedS³: Towards Medical Slow Thinking with Self-Evolved Soft Dual-sided Process Supervision (AAAI 2026 · similarity 0.89)
- CARE: Towards Clinical Accountability in Multi-Modal Medical Reasoning with an Evidence-Grounded Agentic Framework (ICLR 2026 · similarity 0.88)
- MedReasoner: Reinforcement Learning Drives Reasoning Grounding from Clinical Thought to Pixel-Level Precision (AAAI 2026 · similarity 0.87)
- Optimizing Inference-Time Compute for Medical Reasoning via Uncertainty Quantification (ICML 2026 · similarity 0.87)
- MMedAgent-RL: Optimizing Multi-Agent Collaboration for Multimodal Medical Reasoning (ICLR 2026 · similarity 0.86)
- DoAtlas-1: A Causal Compilation Paradigm for Clinical AI (NeurIPS 2026 · similarity 0.86)
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.
