self-supervised learning - Self-supervised learning uses unlabeled data to learn representations by predicting missing parts, enabling robust models without labeled annotations.