Unknown Class

Open-set recognition (OSR) focuses on developing machine learning models capable of identifying data points belonging to unknown classes, a crucial step beyond traditional closed-set classification. Current research emphasizes robust evaluation methodologies that account for class imbalances and explores novel architectures and algorithms, such as bidirectional uncertainty-based active learning and dual contrastive learning with target-aware universums, to improve the identification and handling of unknown classes. These advancements are significant for applications where encountering unseen data is inevitable, such as medical diagnosis, anomaly detection, and robotics, improving the reliability and adaptability of AI systems in real-world scenarios.