arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.04441cs.LGstat.ML
收藏

后验漂移下的领域泛化

Domain Generalization Under Posterior Drift

University of Michigan(密歇根大学);Northwestern University(西北大学);Boston University(波士顿大学)

Yilun Zhu, Naihao Deng, Naichen Shi, Aditya Gangrade, Clayton Scott

AI总结:

本文提出了一种在后验漂移假设下进行领域泛化的决策理论框架,并通过语言和视觉任务的实验验证了其有效性。

摘要:

领域泛化(DG)是将从多个分布(或领域)中获得的标记训练数据泛化到一个新测试领域的问题,该领域没有标记数据。对于DG的主流基准数据集,存在一个在所有领域上表现良好的单一分类器。在本工作中,我们研究了一种根本不同的情形,其中领域满足后验漂移假设,即最优分类器可能在不同领域上有显著变化。我们建立了DG在后验漂移下的决策理论框架,并通过语言和视觉任务的实验探讨了该框架的实践意义。

英文摘要:

Domain generalization (DG) is the problem of generalizing from several distributions (or domains), for which labeled training data are available, to a new test domain for which no labeled data is available. For the prevailing benchmark datasets in DG, there exists a single classifier that performs well across all domains. In this work, we study a fundamentally different regime where the domains satisfy a \emph{posterior drift} assumption, in which the optimal classifier might vary substantially with domain. We establish a decision-theoretic framework for DG under posterior drift, and investigate the practical implications of this framework through experiments on language and vision tasks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑