Neural networks

A neural network builds a flexible function by composing layers of weighted transformations and nonlinear activation functions.

A simple artificial neuron computes

where \(w\) contains weights, \(b\) is a bias and \(\phi\) is an activation function.

Learning

Network parameters are adjusted to reduce a loss function. Backpropagation efficiently computes derivatives of the loss with respect to parameters, and an optimisation algorithm updates them.

Biological applications

Neural networks are used with images, sequences, signals and complex multivariable data. They can represent nonlinear patterns but may require substantial data and careful validation.