跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+549新增论文
+443其中 AI 相关
+62已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-21 12:19 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 18:42 — 2026-11-21 18:42(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-21 12:19最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

29,158 已收录论文24,181 AI 相关8,992 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 5 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-21 12:17(北京时间)
新论文 · v1arXiv:2611.24994

What the model knows but does not say: monitoring medical LLMs through verbalized thoughts

Frederik Hauke, Jan Clusmann, Hanna Kreutzer, Christiane Kuhl, Tianyu Han, Felix Barajas Ordonez, Sven Nebelung, Jakob Nikolas Kather 等 9 位作者

原文摘要

Efforts to keep clinical language models safe have concentrated on what enters and leaves the model, but a manipulated model can return an answer that looks entirely ordinary. Here we show that the model's own internal state, rendered in plain language, exposes manipulations the output conceals. Using a natural-language autoencoder, we translate the activations of a multimodal LLM (Gemma-3-27B) into text and let an input-blind judge read this verbalized internal state (the thoughts channel) and the model's answer separately. Across two clinical attack objectives (omitting a pathology, covertly recommending a drug), two input modalities, and two delivery mechanisms, the thoughts channel caught the injection significantly more often than the answer in all eight injected conditions (p<0.05), while both channels held near-zero false-positive rates on clean inputs; the omission result carried over to Llama-3.3-70B. The thoughts channel also registered unsafe clinical requests as harmful while the model was still reading them, before it decided whether to refuse. None of this requires retraining: an off-the-shelf verbalizer and a general-purpose judge are enough.

cs.CL · cs.AI

新论文 · v1arXiv:2611.23700

Auditable Clinical Support with Decision Agents

Tiancheng Bao, Riga Wu, Jan Wienhold, Sebastian Billig, Patrick Winnersbach, Hsuan-Chih Chen, Akis Giannoukos, Jiayang Li 等 16 位作者

原文摘要

Clinicians reviewing a diagnostic work-up will ask what was considered at each step and why the assessment changed. Agents powered by free text can reach physician-level diagnostic accuracy, but their written reasoning does not always reflect the factors that went into the decision. Here, we present Decision2Care, a decision agent that works up the acutely ill patient without producing any text. A decision model assigns a probability to each permitted EHR action, and fixed rules carry out the most probable one. We evaluated Decision2Care on the records of 546 emergency admissions across 12 diseases. Decision2Care reached a diagnostic accuracy of 86.5%, compared with 86.1% for the MIRA-v2 agent, and required a median of 2.2 s per work-up, compared with 92-100 s. Each work-up tracked which options Decision2Care was considering at each step. In 10 of the 12 diseases, the differential diagnosis changed most at the test on which the guideline bases the diagnosis. Where Decision2Care was most confident after a single work-up, all but four of its diagnoses were correct (98.5%). We traced each of these four errors back to a single imaging report. Three of them were corrected when that report was removed. The same kind of decision was used to select the treatments, and Decision2Care's drug orders were judged guideline-adherent more often than MIRA-v2's. In emergency patients, its prediction of death was comparable to that of three intensivists on the same cases. In five of the eight intensive care cohorts, it exceeded the best of the three 24-h severity scores within 2-6 h of records. Our results show that decision agents can carry out clinical work-ups while recording the options, probabilities, and execution rules behind each action for audit.

cs.AI

版本更新 · v4arXiv:2601.19618

The pretraining domain outweighs the training objective in setting the privacy-utility trade-off of differentially private medical image analysis

Soroosh Tayebi Arasteh, Mina Farajiamiri, Mahshad Lotfinia, Behrus Hinrichs-Puladi, Jonas Bienzeisler, Mohamed Alhaskir, Mirabela Rusu, Christiane Kuhl 等 10 位作者

原文摘要

Differential privacy limits how much each training image can influence a medical imaging model but lowers its diagnostic accuracy. Pretrained weights reduce this loss. It is unknown whether the pretraining objective or domain improves accuracy more. Here we train chest radiograph classifiers under differential privacy from five starting points on 591,884 radiographs from five datasets. We show that supervised pretraining on chest radiographs ranks first in 24 of 25 combinations of dataset and privacy setting. Measured by the area under the receiver operating characteristic curve, its lead over supervised pretraining on natural images is 2.6% without privacy and 15.3% at the strictest budget. From that starting point, changing the domain improves this area 2.2 to 3.7 times more than changing the objective. Private pretraining on chest radiographs lowers this area by about 5% under privacy but outperforms the three starting points without patient data at the strictest budget. Under privacy, supervised chest radiograph pretraining leads in the worst-performing demographic subgroup.

cs.CV · cs.AI · cs.LG

版本更新 · v2arXiv:2605.04039

Safety and accuracy diverge across deployment conditions in clinical large language models

Sebastian Wind, Tri-Thien Nguyen, Jeta Sopa, Mahshad Lotfinia, Sebastian Bickelhaup, Michael Uder, Harald Köstler, Gerhard Wellein 等 12 位作者

原文摘要

Clinical large language models (LLMs) are commonly scaled by increasing model size, context length, retrieval complexity, or inference-time compute, on the assumption that higher accuracy implies safer behavior. In medicine, where a few confident or evidence-contradicting errors can outweigh average performance, this assumption may be incomplete. We introduce Safety-Focused Evaluation of Scaling (SaFE-Scale) and instantiate it on RadSaFE-200, a 200-question text-based radiology multiple-choice benchmark with clinician-defined clean evidence, conflict evidence, and option-level labels for high-risk error, unsafe answer, and evidence contradiction. Thirty-four local LLMs were evaluated across six deployment conditions spanning closed-book, evidence-grounded, standard and agentic retrieval- augmented generation (RAG), and max-context prompting. Clean clinician-written evidence, used as an upper-bound, produced the strongest joint gain, raising mean accuracy from 73.5% to 94.1% and reducing high-risk error from 12.0% to 2.6% and dangerous overconfidence from 8.0% to 1.6%. Standard and agentic RAG, max-context prompting, and inference-time compute improved selected metrics but did not reproduce this safety profile. Clinically, consequential errors concentrated in a small subset of questions, and variance decomposition placed deployment condition above model family on every safety metric. Within this evaluation, clinical LLM safety did not track accuracy scaling but emerged as a deployment-level property.

cs.CL · cs.AI · cs.LG

版本更新 · v4arXiv:2607.18451

CANDOR: Chance-Calibrated Neighborhood Discordance in Frozen Encoders for Medical Imaging

Soroosh Tayebi Arasteh, Sven Nebelung, Daniel Truhn

原文摘要

In medical imaging, one frozen foundation encoder is often reused for many tasks. The encoder is usually chosen by the area under the receiver operating characteristic curve (AUROC) of a head trained on its features. The AUROC does not measure whether the encoder separates positive from negative images in its feature space. Chance-calibrated neighborhood discordance (CANDOR) measures this separation without training a head. Among images acquired the same way, a positive image is discordant when its k nearest negatives are more similar to it on average than its k nearest positives are. The discordance rate D is the share of discordant positives. When the neighbors come from positive and negative candidate sets of equal size, D has the chance level 1/2 for features without information about the label. With unequal candidate sets, all 11 chest encoders have rates above 1/2 (0.732 to 0.927) on the 12 chest findings of MIMIC-CXR and CheXpert. These rates increase as the finding becomes rarer. With equal sets, the same encoders have rates below 1/2 (0.277 to 0.463). On the chest test split, linear heads miss 0.301 of the discordant positives and 0.165 of the other positives. Across 22 encoders, 605,443 images, and 8 tasks in 7 domains, the best encoder of every task has a rate below 1/2. Because computing D takes one forward pass per image, a hospital can compare candidate encoders before it trains any head.

cs.LG · cs.AI · cs.CL · cs.CV