SAT-RRG：面向放射报告生成的 LLM 引导自适应训练与词元级推拉优化
SAT-RRG: LLM-Guided Self-Adaptive Training for Radiology Report Generation with Token-Level Push-Pull Optimization
把冻结 LLM 当作弱监督裁判，自动定位预测报告与参考报告之间的语义冲突片段，并在这些稀疏关键片段上施加 push-pull 式的词元级自适应梯度惩罚；这一「错误片段定位 + 自适应加权」思路可迁移到图像描述、VQA 等序列生成任务的训练目标设计中。
原文依据 dynamically adjusts token-level gradients based on semantic discrepancy cues derived from a frozen LLM referee
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论放射报告生成器常常产出流畅的文本，却遗漏关键细节，导致局部语义冲突或征象翻转，这类错误需要更强的惩罚。交叉熵（CE）仅仅提高真实词元 y^* 的概率，并不直接压制模型当前选错的词元 \hat y，而且对所有位置一视同仁，因此纠正并未被优先处理。我们提出一种自适应优化框架，依据来自冻结 LLM 裁判的语义差异线索，动态调整词元级梯度。LLM 本身并非本文贡献——它只提供弱监督以触发自适应学习过程。在该框架内：(i) 预测报告与参考报告之间的语义冲突被自动定位，并用“<e>...</e>”标注（仅在训练时使用）；(ii) 在这些稀疏但关键的片段内施加自适应的更强惩罚。更新遵循推拉（push-pull）方案：错误片段被压低，而非错误词元得到强化。更新强度由两个互补信号控制——归一化熵（用于不确定性校准）与焦点式置信度（用于处理过度自信与信心不足的预测）。在 MIMIC-CXR 与 IU-Xray 上，该框架一致地提升了语言指标（BLEU-4、ROUGE-L、CIDEr）与临床指标（RadGraph F1、CheXbert），并且对噪声或不完美的错误标注保持稳健。
英文原文摘要 1471 字符
Radiology report generators often produce fluent text yet miss crucial details, leading to local semantic conflicts or flipped findings that require stronger penalties. **Cross-entropy (CE) merely increases the probability of the ground-truth token y^* without directly suppressing the model's current wrong choice \hat y **, and treats all positions uniformly, so corrections are not prioritized. We introduce a **self-adaptive optimization framework** that dynamically adjusts token-level gradients based on semantic discrepancy cues derived from a frozen LLM referee. The LLM itself is not the contribution--it merely provides weak supervision to trigger the adaptive learning process. Within this framework, (i) semantic conflicts between the predicted and reference reports are **automatically localized** and tagged with "<e>...</e>" (used only during training), and (ii) **adaptive, stronger penalties** are applied within these sparse but critical spans. Updates follow a *push-pull* scheme: error spans are pushed down, while non-error tokens are reinforced. The update strength is governed by two complementary signals--*normalized entropy* (for uncertainty calibration) and *focal-style confidence* (for handling over- and under-confident predictions). On MIMIC-CXR and IU-Xray, our framework consistently improves both language metrics (BLEU-4, ROUGE-L, CIDEr) and clinical metrics (RadGraph F1, CheXbert), and remains robust to noisy or imperfect error tags.