arXiv：cs.AI· Xiaoliang Wang, Zhuo Chen, Chuanyang Zheng, Quanbin Wang, Jiangwei Lao, Yingao Liu, Jiajia Liu, Jinming Cao, Honglin Lin, Zhihong Lu, Hao Jin, Jian Wang, Lijun Wu·· 4 小时前
MedEvo：通过知识导航实现医学推理进化
MedEvo: Medical Reasoning Evolution through Knowledge Navigation
AI 导读
MedEvo 利用医学标签分类体系和错误类型诊断来导航推理轨迹修复与数据选择，仅靠 SFT 就将 Qwen3.5-9B 在七个基准上的平均准确率从 65.57% 提升至 69.12%，在 MedXpertQA-MM 上从 47.40% 提升至 54.50%，超过教师蒸馏 SFT 的 66.93%。
正文
Authors:Xiaoliang Wang, Zhuo Chen, Chuanyang Zheng, Quanbin Wang, Jiangwei Lao, Yingao Liu, Jiajia Liu, Jinming Cao, Honglin Lin, Zhihong Lu, Hao Jin, Jian Wang, Lijun Wu
Abstract:Medical multimodal reasoning models are typically trained on teacher-generated rationales, discarding the valid reasoning in students' own failed trajectories, while general-domain correction methods offer little guidance for errors in image reading, anatomical context, or differential diagnosis. We propose MedEvo, which uses medical knowledge, namely a tag taxonomy and error-type diagnosis, to navigate both trajectory repair and data selection. For each failed rollout, a gold-blind auditor locates the first consequential error and provides a minimal, answer-free hint; the same student retains the valid prefix and regenerates the erroneous suffix. After each round, MedEvo identifies capability gaps across medical tags on a held-out feedback set and prioritizes related questions for new rollouts and repairs. With supervised fine-tuning (SFT) alone, MedEvo raises Qwen3.5-9B mean accuracy across seven benchmarks from 65.57% to 69.12% (47.40% to 54.50% on the multimodal track of MedXpertQA (MedXpertQA-MM)), surpassing teacher-distillation SFT (66.93%). It also raises Qwen3.5-35B and Qwen3.5-397B mean accuracy from 67.91% to 73.04% and from 71.83% to 74.95%, respectively. Our analyses show that failed trajectories contain reasoning worth preserving, that repair success does not imply learning value, and that independent feedback is key to sustained improvement.
| Comments: | Project: this https URL GitHub: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Quantitative Methods (q-bio.QM) |
| Cite as: | arXiv:2610.17690 [cs.AI] |
| (or arXiv:2610.17690v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17690 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zhuo Chen [view email]
[v1]
Tue, 13 Oct 2026 04:11:26 UTC (4,581 KB)
来源：arXiv：cs.AI · arxiv.org