Deep Learning

A subset of machine learning based on artificial neural networks with multiple layers.

How it works

Deep learning models stack many layers of learnable transformations between their inputs and outputs. In each layer, linear transformations (matrix multiplications) are composed with non-linear activation functions to progressively extract more abstract features from the data. Training uses backpropagation to compute gradients and an optimiser like Adam to update weights. The 'deep' in deep learning refers to having many layers — modern LLMs may have 80 to 120 Transformer layers.

Why it matters

Deep learning has achieved superhuman performance across computer vision, speech recognition, protein folding, strategy games, and natural language understanding. Its ability to learn hierarchical representations directly from raw data — without manual feature engineering — is what separates it from prior machine learning paradigms. Virtually every frontier AI system in production today is built on deep learning foundations.