Label Augmentation

Label augmentation enhances the training of machine learning models by modifying or expanding the available labels, addressing challenges like limited labeled data and noisy labels. Current research focuses on various augmentation strategies, including generating pseudo-labels from model predictions, leveraging label semantics to create synthetic labels, and incorporating label relationships (e.g., shared structures in activity recognition). These techniques improve model robustness, particularly in semi-supervised and few-shot learning scenarios, and lead to better generalization across diverse applications, such as medical image analysis, natural language processing, and audio tagging.