Auxiliary Outlier

Auxiliary outlier methods aim to improve out-of-distribution (OOD) detection in machine learning models by incorporating additional, outlier data during training. Current research focuses on effective strategies for selecting and utilizing these auxiliary outliers, including techniques like outlier prototype learning, boosting algorithms (e.g., Hopfield Boosting), and diverse sampling methods. These advancements are crucial for enhancing the robustness and reliability of machine learning systems deployed in real-world scenarios where encountering unseen data is inevitable, improving safety and performance in applications ranging from image classification to safety-critical systems.