arXiv cs.CL· Yuexin Wu, Dayou Yu, Vasile Rus·· 12 小时前AI 评分20
面向医疗问答数据选择的开发调优表示敏感性实证研究
An Empirical Study of Development-Tuned Representation Sensitivity for Medical QA Data Selection
AI 导读
研究提出 Representation-Sensitivity Prioritization（RSP），先按预测熵将候选池限制到 top-4B，再结合隐藏态扰动下答案翻转频率与前二答案 margin 的非翻转收缩来打分。
正文 · 原文
Abstract:We study medical question-answering (QA) data selection when a developer already has questions, labels, and auxiliary explanatory texts but can retain only $B$ examples for a final LoRA update. Representation-Sensitivity Prioritization (RSP) first restricts the candidate bank to its top-$4B$ predictive-entropy examples, then combines the frequency of answer flips with non-flipping contraction of the top-two answer margin under hidden-state perturbations. We evaluate this operational score on five medical QA datasets with google/medgemma-4b-it, $B=128$, and three seed branches. After removing four MedExpQA test questions that also occur in its acquisition pool, dataset-specific development-tuned RSP attains the highest observed five-dataset mean accuracy (53.44%) and Macro-F1 (48.21%) among the evaluated selectors. Relative to random selection within the same entropy-restricted pool, the descriptive differences are +1.49 and +2.10 percentage points, respectively; paired 95% bootstrap intervals include zero. The audit mixes benchmarks with different relationships to documented backbone post-training sources, and the same test cohorts had been used in earlier analyses before the expanded tuning grid was designed. Results are therefore exploratory, modest, and dataset-dependent rather than evidence of statistical superiority. RSP point estimates also exceed an archived no-update reference, but unpinned model revisions and incomplete historical input identifiers limit attribution of that difference to fine-tuning. Because every selected training example includes auxiliary text and no matched answer-only control is evaluated, the experiments also do not identify a causal benefit from auxiliary-text conditioning itself.
| Comments: | 16 pages. Substantially revised GenAI4Health @ NeurIPS 2026 version: updated title, development-tuned question-only selection protocol, experiments, and retrospective audit. Includes retained-artifact ancillary material |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.09684 [cs.CL] |
| (or arXiv:2609.09684v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2609.09684 arXiv-issued DOI via DataCite |
Submission history
From: Yuexin Wu [view email]
[v1]
Wed, 9 Sep 2026 03:58:59 UTC (79 KB)
[v2]
Tue, 20 Oct 2026 11:35:07 UTC (1,049 KB)
来源：arXiv cs.CL · arxiv.org