Deep learning — Glossary

Deep learning stacks many layers of nonlinear transformations so that features useful for a task can be learned automatically from raw or lightly processed inputs. It has driven major advances in speech recognition, computer vision, and natural language processing when enough data and compute are available.

Training usually relies on gradient-based optimization and backpropagation. Architectural choices—convolutions, attention, residual connections—encode assumptions about structure in the data. Deep learning does not obsolete classical ML for every problem; tabular and low-data settings often still favor simpler models.