Noisy Partial Label

Noisy partial label learning (NPLL) tackles the challenge of training machine learning models with datasets where each data point is associated with a set of possible labels, some of which may be incorrect or may not include the true label. Current research focuses on developing robust algorithms, often incorporating techniques like pseudo-labeling, label smoothing, and contrastive learning, to effectively disambiguate noisy labels and improve model accuracy. These methods are applied to various model architectures, including convolutional neural networks and vision transformers, and are evaluated across diverse datasets. The ability to effectively learn from NPLL datasets is crucial for improving the reliability and scalability of machine learning in real-world applications where obtaining perfectly labeled data is often impractical.