All terms
Foundations
Representation Learning
Learning useful internal forms of data automatically rather than hand-engineering them.
Definition
Representation learning is the practice of having a model learn useful internal forms of its data automatically, rather than relying on hand-engineered features. Deep networks build up such representations layer by layer, and self-supervised pretraining produces general-purpose ones from unlabeled data. These learned representations, often vectors in a latent space, transfer well to downstream tasks like classification and search.