English(EN) Is Complex Training Necessary for Long-Tailed OOD Detection? A Re-think from Feature Geometry

新型 HPM 检测器通过简化的特征几何提升 OOD 检测性能

作者 PulseAugur 编辑部 · [1 个来源] · 2026-05-18 03:21

研究人员提出了一种名为超球池化马氏距离 (HPM) 的新型事后检测器，用于长尾分布外 (LT-OOD) 检测。该方法对特征进行归一化，并使用池化、岭正则化的度量，而非特定类别的协方差。在 CIFAR-LT 和 ImageNet-100-LT 数据集上的实验表明，HPM 显著提高了检测精度，在 CIFAR-100-LT 上取得了最先进的 Log Efficiency Score，同时以更低的训练成本保持了高 AUROC。 AI

影响引入了一种更有效的方法来检测分布外数据，有望提高模型在实际应用中的鲁棒性。

排序理由该集群包含一篇详细介绍特定机器学习任务新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 0 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍特定机器学习任务新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, other; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: 112 days old; Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CV TIER_1 English(EN) · Yanhui Gu · 2026-05-18 03:21

长尾 OOD 检测是否需要复杂训练？从特征几何角度再思考

Long-tailed out-of-distribution (LT-OOD) detection is often addressed with specialized training, including auxiliary out-of-distribution (OOD) data, abstention heads, contrastive objectives, energy losses, or gradient-conflict control. We show that these training mechanisms can o…

本文如何被排名

报道来源 [1]

长尾 OOD 检测是否需要复杂训练？从特征几何角度再思考

相关实体

相关话题