arXiv:2605.28021v1 Announce Type: new 
Abstract: Out-of-distribution (OOD) detection is essential for deploying machine learning models in open-world and safety-critical scenarios, where test inputs may deviate from the training distribution and overconfident predictions on unknown samples can lead to unreliable decisions. Outlier Exposure (OE) has emerged as a promising OOD detection paradigm by introducing auxiliary outliers during training to enlarge the margin between in-distribution (ID) and OOD samples. Existing OE-based methods typically enlarge this margin by employing uniform labels to maximize the entropy of OOD samples over ID categories. However, we theoretically show that uniform labels inevitably disregard the relations between OOD samples and ID categories, termed the over-softening effect, leading to a suboptimal margin bound. Our theoretical analysis further reveals that explicitly exploiting such relations can instead yield improved OOD detection performance. Motivated by this insight, we propose \underline{A}daptive Confidence \underline{OE} (AOE), a simple yet effective method that leverages temperature scaling to recalibrate outlier labels. Specifically, AOE generates adaptive soft targets from temperature-scaled model predictions for OOD samples, where the learnable temperature smooths the prediction distribution without fully erasing class-wise relational information. By supervising OOD samples with these adaptive soft targets, AOE preserves the semantic proximity between OOD samples and ID categories while encouraging the softened targets to approach a high-entropy distribution, thereby suppressing overconfident OOD predictions and enlarging the separation margin. Extensive experiments across diverse benchmarks demonstrate the effectiveness of AOE.

تقدم دراسة جديدة AOE، وهي طريقة للكشف عن البيانات خارج التوزيع (OOD) تعيد ضبط تسميات القيم الشاذة لتحسين أداء نماذج التعلم الآلي في السيناريوهات التي قد تختلف فيها مدخلات الاختبار عن بيانات التدريب. تتناول هذه الطريقة قيود الأساليب الحالية للكشف عن القيم الشاذة التي تستخدم تسميات موحدة، مما قد يؤدي إلى أداء كشف دون المستوى بسبب تأثير التخفيف المفرط.

Un nuevo estudio presenta AOE, un método para la detección de datos fuera de distribución (OOD) que recalibra las etiquetas de valores atípicos para mejorar el rendimiento de los modelos de aprendizaje automático en escenarios donde las entradas de prueba pueden diferir de los datos de entrenamiento. Este enfoque aborda las limitaciones de los métodos de exposición a valores atípicos existentes que utilizan etiquetas uniformes, lo que puede llevar a un rendimiento de detección subóptimo debido al efecto de sobre-suavizado.

Une nouvelle étude présente AOE, une méthode de détection des données hors distribution (OOD) qui recalibre les étiquettes des valeurs aberrantes pour améliorer la performance des modèles d'apprentissage automatique dans des scénarios où les données de test peuvent différer des données d'entraînement. Cette approche aborde les limites des méthodes d'exposition aux valeurs aberrantes existantes qui utilisent des étiquettes uniformes, pouvant mener à une performance de détection sous-optimale en raison de l'effet de sur-adoucissement.

A new study introduces AOE, a method for out-of-distribution (OOD) detection that recalibrates outlier labels to enhance the performance of machine learning models in scenarios where test inputs may differ from training data. This approach addresses the limitations of existing outlier exposure methods that use uniform labels, which can lead to suboptimal detection performance due to the over-softening effect.