arXiv：cs.LG· Sazan Mahbub, Caleb N. Ellington, Zhiyuan Li, Yixin Yang, Souvik Kundu, Ben Lengerich, Eric P. Xing·· 7 小时前AI 评分33
RAIL：面向医疗保健任务特定零样本模型的可解释检索增强学习
Retrieval-Augmented Interpretable Learning: Towards Task-Specific Zero-Shot Models in Healthcare
AI 导读
RAIL 是一个概率元学习框架，可从自然语言任务描述和已学预测器记忆中合成系数空间结构，实现在原始诊断特征空间直接生成任务特定可解释模型。该框架支持零样本和少样本临床程序预测并提供特征级解释，其概率形式对检索、模型系数和预测给出不确定性，可将不确定预测或不稳定解释标记出来交由临床人工复核。在长尾临床程序预测任务上，RAIL 改善了低数据条件下的模型生成，并输出检索、不确定性和系数级诊断信息。
正文
Abstract:We introduce Retrieval-Augmented Interpretable Learning (RAIL), a probabilistic meta-learning framework for zero-shot generation of task-specific interpretable models that synthesizes coefficient-space structure from natural-language task descriptions and a memory of previously learned task-specific predictors. RAIL retrieves related source tasks, transfers structure through coefficient space, and generates a new predictor in the original diagnostic-feature space, enabling zero-shot and few-shot clinical procedure prediction with feature-level explanations. Its probabilistic formulation provides uncertainty over retrieval, model coefficients, and predictions, supporting uncertainty-aware deployment: uncertain predictions or unstable explanations can be flagged for additional clinical review rather than treated as automatic decisions. This makes RAIL particularly suited for healthcare settings, where prediction tasks are highly long-tailed, new clinical targets arise frequently, and models must remain inspectable, uncertainty-aware, and compatible with human oversight. Across long-tailed clinical procedure prediction tasks, RAIL improves low-data model generation, benefits from clinically informed task representations, and yields retrieval, uncertainty, and coefficient-level diagnostics that make model behavior more transparent. These results suggest a path toward scalable clinical prediction systems that can adapt to new tasks while preserving interpretability and reliability.
| Comments: | We note that a preliminary, non-archival workshop version of this work is available online under the name RAG-IM. RAIL is the renamed and completed version of the same work. This manuscript supersedes that earlier non-archival workshop version and should be consulted for the current formulation, results, and contributions |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2607.17508 [cs.LG] |
| (or arXiv:2607.17508v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2607.17508 arXiv-issued DOI via DataCite |
Submission history
From: Sazan Mahbub [view email]
[v1]
Mon, 20 Jul 2026 03:31:36 UTC (2,275 KB)
[v2]
Fri, 31 Jul 2026 21:38:34 UTC (2,275 KB)
[v3]
Tue, 6 Oct 2026 07:50:24 UTC (2,333 KB)
来源：arXiv：cs.LG · arxiv.org