Complementary Label

Complementary label learning (CLL) is a weakly supervised machine learning approach that trains classifiers using labels indicating classes a data point *does not* belong to, rather than its true class. Current research focuses on developing robust algorithms that handle noisy or biased complementary labels, often employing techniques like negative learning, risk correction, and data augmentation to improve model accuracy and generalization. This area is significant because it reduces the need for expensive, fully annotated datasets, impacting various applications, including medical image analysis and domain adaptation, where obtaining complete labels is challenging or impractical.