arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2405.03052cs.LG
收藏

从统计检验理论视角看分布外识别

A View on Out-of-Distribution Identification from a Statistical Testing Theory Perspective

The Alan Turing Institute(阿兰·图灵研究所)

Alberto Caron, Chris Hicks, Vasilios Mavroudis

AI总结:

本文从统计检验理论重新表述监督与无监督场景下的 OOD 检测问题,给出可识别条件,并分析基于 Wasserstein 距离的 OOD 检验的收敛保证及实证结果。

摘要:

我们研究在监督和无监督学习场景中于测试时高效检测分布外(Out-of-Distribution,OOD)样本的问题。尽管机器学习模型通常在训练数据与测试数据来自同一分布的假设下进行训练,但在现实环境中情况往往并非如此,因此在部署时可靠地检测分布偏移至关重要。我们借助统计检验的视角重新表述 OOD 问题,随后讨论使该 OOD 问题在统计意义上可识别的条件。基于这一框架,我们研究一种基于 Wasserstein 距离的 OOD 检验的收敛保证,并提供一项简单的实证评估。

英文摘要:

We study the problem of efficiently detecting Out-of-Distribution (OOD) samples at test time in supervised and unsupervised learning contexts. While ML models are typically trained under the assumption that training and test data stem from the same distribution, this is often not the case in realistic settings, thus reliably detecting distribution shifts is crucial at deployment. We re-formulate the OOD problem under the lenses of statistical testing and then discuss conditions that render the OOD problem identifiable in statistical terms. Building on this framework, we study convergence guarantees of an OOD test based on the Wasserstein distance, and provide a simple empirical evaluation.

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。

↑