Unsupervised Data Augmentation is a semi-supervised learning technique that applies data augmentation to both labeled and unlabeled data. It leverages confidence-based pseudo-labels for unlabeled samples and consistency regularization. UDA can significantly improve performance when labels are scarce.