pseudo-labels

Pseudo-labels are labels generated from the model’s own predictions on unlabelled data. They are often used in semi-supervised learning strategies to increase the effective size of training data, by treating high-confidence predictions as if they were true labels.

15 papers