Consistency Regularization

A core technique in semi-supervised learning and robust training based on the smoothness assumption. It constrains model outputs to remain consistent when different perturbations are applied to the same input sample, improving generalization without requiring additional labeled data.