Foundations · MLD-0289

Deep learning

Machine learning using neural networks with multiple layers of learned representations.

Expanded guide · editorial review 2026-11-01. Original definitions, examples, notes and citations are preserved below. New examples are illustrative, not benchmark results.

Prerequisites

How it works

Deep learning composes learned transformations. Earlier layers produce representations that later layers reuse. Nonlinear activations allow a network to express relationships that a chain of linear transformations cannot. The architecture supplies an inductive bias; examples and optimization determine the learned parameters.

Worked example

In an illustrative image classifier, a convolution transforms a photograph into feature maps, later layers combine them, and a final classifier produces category scores. A training error flows backward through the composition to update the filters. No individual filter has to correspond to a human-named concept.

A useful equation

For a simple two-layer network: h = activation(W1 x + b1); prediction = W2 h + b2. Backpropagation applies the chain rule to compute parameter gradients.

What to test

Compare a simple baseline with the network under the same split and metric. Vary data quantity and random seed; inspect performance on rare categories rather than relying only on an average.