以最少微调数据实现高效医学推理
Towards Efficient Medical Reasoning with Minimal Fine-Tuning Data
对做医学 VLM 适配的人，这篇给出了一个可直接复用的数据筛选准则：用样本梯度影响与难度的二维象限联合打分，只保留「高难度-高影响」样本，从而在 1% 数据下逼近全量微调效果，可迁移到任意 VLM 的领域 SFT 数据蒸馏流程。
原文依据 we propose a data selection strategy, Difficulty-Influence Quadrant (DIQ), which prioritizes samples in the "high-difficulty-high-influence" quadrant
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论语言骨干网络的监督微调（SFT）在将视觉语言模型（VLM）适配到医学推理等专业领域时起着关键作用。然而，现有的 SFT 实践往往依赖未经筛选的文本数据集，其中包含冗余与低质量样本，导致高昂的计算开销，并在复杂临床场景下性能欠佳。尽管现有方法尝试依据由知识与推理复杂度定义的样本难度来筛选数据以缓解该问题，但它们忽略了每个样本体现在其梯度中的优化效用。有趣的是，我们发现仅依赖基于梯度的影响会偏向那些易于优化、能引起较大参数偏移但缺乏深层推理链的样本，而仅依赖难度则会选出噪声大或过于复杂的文本案例，无法引导稳定的优化。基于这一观察，我们提出一种数据选择策略——难度-影响象限（DIQ），它优先选择处于「高难度-高影响」象限的样本，以在复杂临床推理与显著梯度影响之间取得平衡。这使得 VLM 能够以最少的微调数据实现高效的医学推理。此外，人工评估与 LLM-as-a-judge 评估表明，DIQ 选出的子集具有更高的数据质量，并能生成在鉴别诊断、安全检查与证据引用方面更贴合专家实践的临床推理，因为 DIQ 强调那些能够培养类专家推理模式的样本。在医学推理基准上的大量实验表明，DIQ 使仅在 1% 精选数据上微调的 VLM 骨干网络即可达到全量数据集的性能，而使用 10% 数据时则持续优于基线方法，凸显了有原则的数据选择相对于暴力式规模扩展的优越性。代码已发布于 https://github.com/mihara-bot/DIQ。
英文原文摘要 1913 字符
Supervised Fine-Tuning (SFT) of the language backbone plays a pivotal role in adapting Vision-Language Models (VLMs) to specialized domains such as medical reasoning. However, existing SFT practices often rely on unfiltered textual datasets that contain redundant and low-quality samples, leading to substantial computational costs and suboptimal performance in complex clinical scenarios. Although existing methods attempt to alleviate this problem by selecting data based on sample difficulty, defined by knowledge and reasoning complexity, they overlook each sample's optimization utility reflected in its gradient. Interestingly, we find that gradient-based influence alone favors easy-to-optimize samples that cause large parameter shifts but lack deep reasoning chains, while difficulty alone selects noisy or overly complex textual cases that fail to guide stable optimization. Based on this observation, we propose a data selection strategy, Difficulty-Influence Quadrant (DIQ), which prioritizes samples in the "high-difficulty-high-influence" quadrant to balance complex clinical reasoning with substantial gradient influence. This enables efficient medical reasoning for VLMs with minimal fine-tuning data. Furthermore, Human and LLM-as-a-judge evaluations show that DIQ-selected subsets demonstrate higher data quality and generate clinical reasoning that is more aligned with expert practices in differential diagnosis, safety check, and evidence citation, as DIQ emphasizes samples that foster expert-like reasoning patterns. Extensive experiments on medical reasoning benchmarks demonstrate that DIQ enables VLM backbones fine-tuned on only 1% of selected data to match full-dataset performance, while using 10% consistently outperforms baseline methods, highlighting the superiority of principled data selection over brute-force scaling. The code is available at https://github.com/mihara-bot/DIQ.