RankOOD：基于类别排序的分布外检测
RankOOD - Class Ranking-based Out-of-Distribution Detection
把基础模型偏好对齐中常用的 Plackett-Luce 排序损失迁移到 OOD 检测：先用交叉熵分类器为每个 ID 类提取一个固定的类别排序，再用 PL 损失训练模型预测该排序，从而用「是否遵循排序」作为 OOD 分数，可直接替换现有分类器的置信度打分。
原文依据 An OOD example may get assigned with high probability to an ID example, but the probability of it respecting the ranking classification is likely to be small.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论我们提出 RankOOD，一种基于排序的分布外（OOD）检测方法，其核心是用 Plackett-Luce 损失训练模型，该损失目前被广泛用于基础模型的偏好对齐任务。我们的方法基于如下观察：对于使用交叉熵损失训练的深度学习模型，分布内（ID）类别的预测会为每一次 ID 类别预测诱导出一种排序模式。RankOOD 框架将该观察形式化：首先用一个初始分类器为每个类别提取一个排序列表，然后使用 Plackett-Luce 损失进行第二轮训练，其中类别排序（对每个类别而言是一个固定的置换）是被预测的变量。一个 OOD 样本可能被以较高概率分配给某个 ID 类别，但它遵循该排序分类的概率很可能很小。RankOOD 在近 OOD 的 TinyImageNet 评测基准上取得了 SOTA 性能，将 FPR95 降低了 4.3%。
英文原文摘要 960 字符
We propose RankOOD, a rank-based Out-of-Distribution (OOD) detection approach based on training a model with the Placket-Luce loss, which is now extensively used for preference alignment tasks in foundational models. Our approach is based on the insight that with a deep learning model trained using the Cross Entropy Loss, in-distribution (ID) class prediction induces a ranking pattern for each ID class prediction. The RankOOD framework formalizes the insight by first extracting a rank list for each class using an initial classifier and then uses another round of training with the Plackett-Luce loss, where the class rank, a fixed permutation for each class, is the predicted variable. An OOD example may get assigned with high probability to an ID example, but the probability of it respecting the ranking classification is likely to be small. RankOOD, achieves SOTA performance on the near-ODD TinyImageNet evaluation benchmark, reducing FPR95 by 4.3%.