Learns representations from unlabeled data by leveraging intrinsic data properties. Reduces dependency on labeled data and improves scalability.
