绕过传输计划：基于最优传输动态重加权的分布外检测
Bypassing the Transport Plan: Dynamic Reweighting for Out-of-Distribution Detection with Optimal Transport
把 OOD 检测建模为半不平衡最优传输，并利用源分布权重的动态重加权直接得到伪 OOD 分数，绕开求解传输计划；做开放集半监督训练或需要无标签 OOD 打分时可直接复用这一打分模块。
原文依据 we start by formulating OOD detection as a semi-unbalanced optimal transport (SemiUOT) problem
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论半监督学习（SSL）通过同时利用有限的标注数据和大量无标注数据取得了显著进展。然而，无标注数据集往往包含来自未知类别的分布外（OOD）样本，这会在开放集半监督学习场景中导致性能下降。当前的 OOD 检测方法受制于缺少带标注的 OOD 样本。尽管最优传输（OT）已被证明能够有效地为监督学习提供伪 OOD 分数，但它仍面临两个主要挑战，即寻找最优传输计划这一不可避免的问题，以及稠密解所导致的不够可靠的 OOD 分数。为克服这些局限，我们提出了一种新的开放集 OOD 检测模型 DREW，它利用动态重加权（Dynamic REWeighting）方法进行基于 OT 的 OOD 检测。具体而言，我们首先将 OOD 检测建模为半不平衡最优传输（SemiUOT）问题。所提出的 DREW 模型能够动态地将 SemiUOT 转化为经典 OT 形式，进而从新的源分布权重中直接获得伪 OOD 分数。与现有的基于 OT 的方法相反，DREW 提供了具有理论依据且更准确的伪 OOD 分数，同时避免了直接计算传输计划。实验结果在准确性和效率两方面均证明了 DREW 的优越性。我们开展了大量分析性实验，以阐明各个组件的性质。
英文原文摘要 1483 字符
Semi-supervised learning (SSL) has achieved remarkable progress by leveraging both limited labeled data and abundant unlabeled data. However, unlabeled datasets often contain out-of-distribution (OOD) samples from unknown classes, which can lead to performance degradation in open-set SSL scenarios. Current OOD detection methods are constrained by the absence of labeled OOD samples. Although optimal transport (OT) has proven to be effective to provide pseudo OOD scores for supervised learning, it still faces two main challenges, i.e., the unavoidable problem of finding the optimal transport plan and the unreliable OOD score caused by dense solutions. To overcome thess limitations, we propose a novel open-set OOD detection model named **DREW**, which leverages **D**ynamic **REW**eighting approach for OT-based OOD detection. Specifically, we start by formulating OOD detection as a semi-unbalanced optimal transport (SemiUOT) problem. The proposed DREW model can dynamically transform SemiUOT into the classical OT formula and then directly obtain the pseudo OOD score from the new source distribution weights. Contrary to existing OT-based methods, DREW provides theoretically grounded and more accurate pseudo OOD scores, while avoiding the direct computation of the transport plan. Empirical results demonstrate the superiority of DREW in terms of both accuracy and efficiency. Extensive analytical experiments are conducted to elucidate the properties of each component.